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, 2014 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 2015 Nexenta Systems, Inc. All rights reserved.
29 /* Portions Copyright 2010 Robert Milkowski */
32 #include <sys/zfs_context.h>
33 #include <sys/dmu_objset.h>
34 #include <sys/dsl_dir.h>
35 #include <sys/dsl_dataset.h>
36 #include <sys/dsl_prop.h>
37 #include <sys/dsl_pool.h>
38 #include <sys/dsl_synctask.h>
39 #include <sys/dsl_deleg.h>
40 #include <sys/dnode.h>
43 #include <sys/dmu_tx.h>
46 #include <sys/dmu_impl.h>
47 #include <sys/zfs_ioctl.h>
49 #include <sys/zfs_onexit.h>
50 #include <sys/dsl_destroy.h>
53 * Needed to close a window in dnode_move() that allows the objset to be freed
54 * before it can be safely accessed.
61 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
71 dmu_objset_spa(objset_t *os)
77 dmu_objset_zil(objset_t *os)
83 dmu_objset_pool(objset_t *os)
87 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
88 return (ds->ds_dir->dd_pool);
90 return (spa_get_dsl(os->os_spa));
94 dmu_objset_ds(objset_t *os)
96 return (os->os_dsl_dataset);
100 dmu_objset_type(objset_t *os)
102 return (os->os_phys->os_type);
106 dmu_objset_name(objset_t *os, char *buf)
108 dsl_dataset_name(os->os_dsl_dataset, buf);
112 dmu_objset_id(objset_t *os)
114 dsl_dataset_t *ds = os->os_dsl_dataset;
116 return (ds ? ds->ds_object : 0);
120 dmu_objset_syncprop(objset_t *os)
122 return (os->os_sync);
126 dmu_objset_logbias(objset_t *os)
128 return (os->os_logbias);
132 checksum_changed_cb(void *arg, uint64_t newval)
137 * Inheritance should have been done by now.
139 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
141 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
145 compression_changed_cb(void *arg, uint64_t newval)
150 * Inheritance and range checking should have been done by now.
152 ASSERT(newval != ZIO_COMPRESS_INHERIT);
154 os->os_compress = zio_compress_select(newval, ZIO_COMPRESS_ON_VALUE);
158 copies_changed_cb(void *arg, uint64_t newval)
163 * Inheritance and range checking should have been done by now.
166 ASSERT(newval <= spa_max_replication(os->os_spa));
168 os->os_copies = newval;
172 dedup_changed_cb(void *arg, uint64_t newval)
175 spa_t *spa = os->os_spa;
176 enum zio_checksum checksum;
179 * Inheritance should have been done by now.
181 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
183 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
185 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
186 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
190 primary_cache_changed_cb(void *arg, uint64_t newval)
195 * Inheritance and range checking should have been done by now.
197 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
198 newval == ZFS_CACHE_METADATA);
200 os->os_primary_cache = newval;
204 secondary_cache_changed_cb(void *arg, uint64_t newval)
209 * Inheritance and range checking should have been done by now.
211 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
212 newval == ZFS_CACHE_METADATA);
214 os->os_secondary_cache = newval;
218 sync_changed_cb(void *arg, uint64_t newval)
223 * Inheritance and range checking should have been done by now.
225 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
226 newval == ZFS_SYNC_DISABLED);
228 os->os_sync = newval;
230 zil_set_sync(os->os_zil, newval);
234 redundant_metadata_changed_cb(void *arg, uint64_t newval)
239 * Inheritance and range checking should have been done by now.
241 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
242 newval == ZFS_REDUNDANT_METADATA_MOST);
244 os->os_redundant_metadata = newval;
248 logbias_changed_cb(void *arg, uint64_t newval)
252 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
253 newval == ZFS_LOGBIAS_THROUGHPUT);
254 os->os_logbias = newval;
256 zil_set_logbias(os->os_zil, newval);
260 recordsize_changed_cb(void *arg, uint64_t newval)
264 os->os_recordsize = newval;
268 dmu_objset_byteswap(void *buf, size_t size)
270 objset_phys_t *osp = buf;
272 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
273 dnode_byteswap(&osp->os_meta_dnode);
274 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
275 osp->os_type = BSWAP_64(osp->os_type);
276 osp->os_flags = BSWAP_64(osp->os_flags);
277 if (size == sizeof (objset_phys_t)) {
278 dnode_byteswap(&osp->os_userused_dnode);
279 dnode_byteswap(&osp->os_groupused_dnode);
284 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
290 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
292 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
293 os->os_dsl_dataset = ds;
296 if (!BP_IS_HOLE(os->os_rootbp)) {
297 arc_flags_t aflags = ARC_FLAG_WAIT;
299 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
300 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
302 if (DMU_OS_IS_L2CACHEABLE(os))
303 aflags |= ARC_FLAG_L2CACHE;
304 if (DMU_OS_IS_L2COMPRESSIBLE(os))
305 aflags |= ARC_FLAG_L2COMPRESS;
307 dprintf_bp(os->os_rootbp, "reading %s", "");
308 err = arc_read(NULL, spa, os->os_rootbp,
309 arc_getbuf_func, &os->os_phys_buf,
310 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
312 kmem_free(os, sizeof (objset_t));
313 /* convert checksum errors into IO errors */
315 err = SET_ERROR(EIO);
319 /* Increase the blocksize if we are permitted. */
320 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
321 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
322 arc_buf_t *buf = arc_buf_alloc(spa,
323 sizeof (objset_phys_t), &os->os_phys_buf,
325 bzero(buf->b_data, sizeof (objset_phys_t));
326 bcopy(os->os_phys_buf->b_data, buf->b_data,
327 arc_buf_size(os->os_phys_buf));
328 (void) arc_buf_remove_ref(os->os_phys_buf,
330 os->os_phys_buf = buf;
333 os->os_phys = os->os_phys_buf->b_data;
334 os->os_flags = os->os_phys->os_flags;
336 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
337 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
338 os->os_phys_buf = arc_buf_alloc(spa, size,
339 &os->os_phys_buf, ARC_BUFC_METADATA);
340 os->os_phys = os->os_phys_buf->b_data;
341 bzero(os->os_phys, size);
345 * Note: the changed_cb will be called once before the register
346 * func returns, thus changing the checksum/compression from the
347 * default (fletcher2/off). Snapshots don't need to know about
348 * checksum/compression/copies.
351 err = dsl_prop_register(ds,
352 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
353 primary_cache_changed_cb, os);
355 err = dsl_prop_register(ds,
356 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
357 secondary_cache_changed_cb, os);
359 if (!dsl_dataset_is_snapshot(ds)) {
361 err = dsl_prop_register(ds,
362 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
363 checksum_changed_cb, os);
366 err = dsl_prop_register(ds,
367 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
368 compression_changed_cb, os);
371 err = dsl_prop_register(ds,
372 zfs_prop_to_name(ZFS_PROP_COPIES),
373 copies_changed_cb, os);
376 err = dsl_prop_register(ds,
377 zfs_prop_to_name(ZFS_PROP_DEDUP),
378 dedup_changed_cb, os);
381 err = dsl_prop_register(ds,
382 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
383 logbias_changed_cb, os);
386 err = dsl_prop_register(ds,
387 zfs_prop_to_name(ZFS_PROP_SYNC),
388 sync_changed_cb, os);
391 err = dsl_prop_register(ds,
393 ZFS_PROP_REDUNDANT_METADATA),
394 redundant_metadata_changed_cb, os);
397 err = dsl_prop_register(ds,
398 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
399 recordsize_changed_cb, os);
403 VERIFY(arc_buf_remove_ref(os->os_phys_buf,
405 kmem_free(os, sizeof (objset_t));
409 /* It's the meta-objset. */
410 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
411 os->os_compress = ZIO_COMPRESS_LZJB;
412 os->os_copies = spa_max_replication(spa);
413 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
414 os->os_dedup_verify = B_FALSE;
415 os->os_logbias = ZFS_LOGBIAS_LATENCY;
416 os->os_sync = ZFS_SYNC_STANDARD;
417 os->os_primary_cache = ZFS_CACHE_ALL;
418 os->os_secondary_cache = ZFS_CACHE_ALL;
421 if (ds == NULL || !dsl_dataset_is_snapshot(ds))
422 os->os_zil_header = os->os_phys->os_zil_header;
423 os->os_zil = zil_alloc(os, &os->os_zil_header);
425 for (i = 0; i < TXG_SIZE; i++) {
426 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
427 offsetof(dnode_t, dn_dirty_link[i]));
428 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
429 offsetof(dnode_t, dn_dirty_link[i]));
431 list_create(&os->os_dnodes, sizeof (dnode_t),
432 offsetof(dnode_t, dn_link));
433 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
434 offsetof(dmu_buf_impl_t, db_link));
436 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
437 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
438 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
440 DMU_META_DNODE(os) = dnode_special_open(os,
441 &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT,
443 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
444 DMU_USERUSED_DNODE(os) = dnode_special_open(os,
445 &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT,
446 &os->os_userused_dnode);
447 DMU_GROUPUSED_DNODE(os) = dnode_special_open(os,
448 &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT,
449 &os->os_groupused_dnode);
457 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
461 mutex_enter(&ds->ds_opening_lock);
462 if (ds->ds_objset == NULL) {
464 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
465 ds, dsl_dataset_get_blkptr(ds), &os);
468 mutex_enter(&ds->ds_lock);
469 ASSERT(ds->ds_objset == NULL);
471 mutex_exit(&ds->ds_lock);
474 *osp = ds->ds_objset;
475 mutex_exit(&ds->ds_opening_lock);
480 * Holds the pool while the objset is held. Therefore only one objset
481 * can be held at a time.
484 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
490 err = dsl_pool_hold(name, tag, &dp);
493 err = dsl_dataset_hold(dp, name, tag, &ds);
495 dsl_pool_rele(dp, tag);
499 err = dmu_objset_from_ds(ds, osp);
501 dsl_dataset_rele(ds, tag);
502 dsl_pool_rele(dp, tag);
509 * dsl_pool must not be held when this is called.
510 * Upon successful return, there will be a longhold on the dataset,
511 * and the dsl_pool will not be held.
514 dmu_objset_own(const char *name, dmu_objset_type_t type,
515 boolean_t readonly, void *tag, objset_t **osp)
521 err = dsl_pool_hold(name, FTAG, &dp);
524 err = dsl_dataset_own(dp, name, tag, &ds);
526 dsl_pool_rele(dp, FTAG);
530 err = dmu_objset_from_ds(ds, osp);
531 dsl_pool_rele(dp, FTAG);
533 dsl_dataset_disown(ds, tag);
534 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
535 dsl_dataset_disown(ds, tag);
536 return (SET_ERROR(EINVAL));
537 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
538 dsl_dataset_disown(ds, tag);
539 return (SET_ERROR(EROFS));
545 dmu_objset_rele(objset_t *os, void *tag)
547 dsl_pool_t *dp = dmu_objset_pool(os);
548 dsl_dataset_rele(os->os_dsl_dataset, tag);
549 dsl_pool_rele(dp, tag);
553 * When we are called, os MUST refer to an objset associated with a dataset
554 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
555 * == tag. We will then release and reacquire ownership of the dataset while
556 * holding the pool config_rwlock to avoid intervening namespace or ownership
559 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
560 * release the hold on its dataset and acquire a new one on the dataset of the
561 * same name so that it can be partially torn down and reconstructed.
564 dmu_objset_refresh_ownership(objset_t *os, void *tag)
567 dsl_dataset_t *ds, *newds;
568 char name[MAXNAMELEN];
570 ds = os->os_dsl_dataset;
571 VERIFY3P(ds, !=, NULL);
572 VERIFY3P(ds->ds_owner, ==, tag);
573 VERIFY(dsl_dataset_long_held(ds));
575 dsl_dataset_name(ds, name);
576 dp = dmu_objset_pool(os);
577 dsl_pool_config_enter(dp, FTAG);
578 dmu_objset_disown(os, tag);
579 VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
580 VERIFY3P(newds, ==, os->os_dsl_dataset);
581 dsl_pool_config_exit(dp, FTAG);
585 dmu_objset_disown(objset_t *os, void *tag)
587 dsl_dataset_disown(os->os_dsl_dataset, tag);
591 dmu_objset_evict_dbufs(objset_t *os)
595 mutex_enter(&os->os_lock);
597 /* process the mdn last, since the other dnodes have holds on it */
598 list_remove(&os->os_dnodes, DMU_META_DNODE(os));
599 list_insert_tail(&os->os_dnodes, DMU_META_DNODE(os));
602 * Find the first dnode with holds. We have to do this dance
603 * because dnode_add_ref() only works if you already have a
604 * hold. If there are no holds then it has no dbufs so OK to
607 for (dn = list_head(&os->os_dnodes);
608 dn && !dnode_add_ref(dn, FTAG);
609 dn = list_next(&os->os_dnodes, dn))
613 dnode_t *next_dn = dn;
616 next_dn = list_next(&os->os_dnodes, next_dn);
617 } while (next_dn && !dnode_add_ref(next_dn, FTAG));
619 mutex_exit(&os->os_lock);
620 dnode_evict_dbufs(dn);
621 dnode_rele(dn, FTAG);
622 mutex_enter(&os->os_lock);
625 mutex_exit(&os->os_lock);
629 dmu_objset_evict(objset_t *os)
631 dsl_dataset_t *ds = os->os_dsl_dataset;
633 for (int t = 0; t < TXG_SIZE; t++)
634 ASSERT(!dmu_objset_is_dirty(os, t));
637 if (!dsl_dataset_is_snapshot(ds)) {
638 VERIFY0(dsl_prop_unregister(ds,
639 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
640 checksum_changed_cb, os));
641 VERIFY0(dsl_prop_unregister(ds,
642 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
643 compression_changed_cb, os));
644 VERIFY0(dsl_prop_unregister(ds,
645 zfs_prop_to_name(ZFS_PROP_COPIES),
646 copies_changed_cb, os));
647 VERIFY0(dsl_prop_unregister(ds,
648 zfs_prop_to_name(ZFS_PROP_DEDUP),
649 dedup_changed_cb, os));
650 VERIFY0(dsl_prop_unregister(ds,
651 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
652 logbias_changed_cb, os));
653 VERIFY0(dsl_prop_unregister(ds,
654 zfs_prop_to_name(ZFS_PROP_SYNC),
655 sync_changed_cb, os));
656 VERIFY0(dsl_prop_unregister(ds,
657 zfs_prop_to_name(ZFS_PROP_REDUNDANT_METADATA),
658 redundant_metadata_changed_cb, os));
659 VERIFY0(dsl_prop_unregister(ds,
660 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
661 recordsize_changed_cb, os));
663 VERIFY0(dsl_prop_unregister(ds,
664 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
665 primary_cache_changed_cb, os));
666 VERIFY0(dsl_prop_unregister(ds,
667 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
668 secondary_cache_changed_cb, os));
674 dmu_objset_evict_dbufs(os);
676 dnode_special_close(&os->os_meta_dnode);
677 if (DMU_USERUSED_DNODE(os)) {
678 dnode_special_close(&os->os_userused_dnode);
679 dnode_special_close(&os->os_groupused_dnode);
681 zil_free(os->os_zil);
683 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
685 VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
688 * This is a barrier to prevent the objset from going away in
689 * dnode_move() until we can safely ensure that the objset is still in
690 * use. We consider the objset valid before the barrier and invalid
693 rw_enter(&os_lock, RW_READER);
696 mutex_destroy(&os->os_lock);
697 mutex_destroy(&os->os_obj_lock);
698 mutex_destroy(&os->os_user_ptr_lock);
699 kmem_free(os, sizeof (objset_t));
703 dmu_objset_snap_cmtime(objset_t *os)
705 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
708 /* called from dsl for meta-objset */
710 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
711 dmu_objset_type_t type, dmu_tx_t *tx)
716 ASSERT(dmu_tx_is_syncing(tx));
719 VERIFY0(dmu_objset_from_ds(ds, &os));
721 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
723 mdn = DMU_META_DNODE(os);
725 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
726 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
729 * We don't want to have to increase the meta-dnode's nlevels
730 * later, because then we could do it in quescing context while
731 * we are also accessing it in open context.
733 * This precaution is not necessary for the MOS (ds == NULL),
734 * because the MOS is only updated in syncing context.
735 * This is most fortunate: the MOS is the only objset that
736 * needs to be synced multiple times as spa_sync() iterates
737 * to convergence, so minimizing its dn_nlevels matters.
743 * Determine the number of levels necessary for the meta-dnode
744 * to contain DN_MAX_OBJECT dnodes.
746 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
747 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
748 DN_MAX_OBJECT * sizeof (dnode_phys_t))
751 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
752 mdn->dn_nlevels = levels;
755 ASSERT(type != DMU_OST_NONE);
756 ASSERT(type != DMU_OST_ANY);
757 ASSERT(type < DMU_OST_NUMTYPES);
758 os->os_phys->os_type = type;
759 if (dmu_objset_userused_enabled(os)) {
760 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
761 os->os_flags = os->os_phys->os_flags;
764 dsl_dataset_dirty(ds, tx);
769 typedef struct dmu_objset_create_arg {
770 const char *doca_name;
772 void (*doca_userfunc)(objset_t *os, void *arg,
773 cred_t *cr, dmu_tx_t *tx);
775 dmu_objset_type_t doca_type;
777 } dmu_objset_create_arg_t;
781 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
783 dmu_objset_create_arg_t *doca = arg;
784 dsl_pool_t *dp = dmu_tx_pool(tx);
789 if (strchr(doca->doca_name, '@') != NULL)
790 return (SET_ERROR(EINVAL));
792 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
796 dsl_dir_rele(pdd, FTAG);
797 return (SET_ERROR(EEXIST));
799 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
801 dsl_dir_rele(pdd, FTAG);
807 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
809 dmu_objset_create_arg_t *doca = arg;
810 dsl_pool_t *dp = dmu_tx_pool(tx);
818 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
820 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
821 doca->doca_cred, tx);
823 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
824 bp = dsl_dataset_get_blkptr(ds);
825 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
826 ds, bp, doca->doca_type, tx);
828 if (doca->doca_userfunc != NULL) {
829 doca->doca_userfunc(os, doca->doca_userarg,
830 doca->doca_cred, tx);
833 spa_history_log_internal_ds(ds, "create", tx, "");
834 dsl_dataset_rele(ds, FTAG);
835 dsl_dir_rele(pdd, FTAG);
839 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
840 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
842 dmu_objset_create_arg_t doca;
844 doca.doca_name = name;
845 doca.doca_cred = CRED();
846 doca.doca_flags = flags;
847 doca.doca_userfunc = func;
848 doca.doca_userarg = arg;
849 doca.doca_type = type;
851 return (dsl_sync_task(name,
852 dmu_objset_create_check, dmu_objset_create_sync, &doca,
853 5, ZFS_SPACE_CHECK_NORMAL));
856 typedef struct dmu_objset_clone_arg {
857 const char *doca_clone;
858 const char *doca_origin;
860 } dmu_objset_clone_arg_t;
864 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
866 dmu_objset_clone_arg_t *doca = arg;
870 dsl_dataset_t *origin;
871 dsl_pool_t *dp = dmu_tx_pool(tx);
873 if (strchr(doca->doca_clone, '@') != NULL)
874 return (SET_ERROR(EINVAL));
876 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
880 dsl_dir_rele(pdd, FTAG);
881 return (SET_ERROR(EEXIST));
884 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
887 dsl_dir_rele(pdd, FTAG);
888 return (SET_ERROR(EDQUOT));
890 dsl_dir_rele(pdd, FTAG);
892 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
896 /* You can only clone snapshots, not the head datasets. */
897 if (!dsl_dataset_is_snapshot(origin)) {
898 dsl_dataset_rele(origin, FTAG);
899 return (SET_ERROR(EINVAL));
901 dsl_dataset_rele(origin, FTAG);
907 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
909 dmu_objset_clone_arg_t *doca = arg;
910 dsl_pool_t *dp = dmu_tx_pool(tx);
913 dsl_dataset_t *origin, *ds;
915 char namebuf[MAXNAMELEN];
917 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
918 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
920 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
921 doca->doca_cred, tx);
923 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
924 dsl_dataset_name(origin, namebuf);
925 spa_history_log_internal_ds(ds, "clone", tx,
926 "origin=%s (%llu)", namebuf, origin->ds_object);
927 dsl_dataset_rele(ds, FTAG);
928 dsl_dataset_rele(origin, FTAG);
929 dsl_dir_rele(pdd, FTAG);
933 dmu_objset_clone(const char *clone, const char *origin)
935 dmu_objset_clone_arg_t doca;
937 doca.doca_clone = clone;
938 doca.doca_origin = origin;
939 doca.doca_cred = CRED();
941 return (dsl_sync_task(clone,
942 dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
943 5, ZFS_SPACE_CHECK_NORMAL));
947 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
950 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
951 nvlist_t *snaps = fnvlist_alloc();
953 fnvlist_add_boolean(snaps, longsnap);
955 err = dsl_dataset_snapshot(snaps, NULL, NULL);
961 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
965 while (dn = list_head(list)) {
966 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
967 ASSERT(dn->dn_dbuf->db_data_pending);
969 * Initialize dn_zio outside dnode_sync() because the
970 * meta-dnode needs to set it ouside dnode_sync().
972 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
975 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
976 list_remove(list, dn);
979 (void) dnode_add_ref(dn, newlist);
980 list_insert_tail(newlist, dn);
989 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
991 blkptr_t *bp = zio->io_bp;
993 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
995 ASSERT(!BP_IS_EMBEDDED(bp));
996 ASSERT3P(bp, ==, os->os_rootbp);
997 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
998 ASSERT0(BP_GET_LEVEL(bp));
1001 * Update rootbp fill count: it should be the number of objects
1002 * allocated in the object set (not counting the "special"
1003 * objects that are stored in the objset_phys_t -- the meta
1004 * dnode and user/group accounting objects).
1007 for (int i = 0; i < dnp->dn_nblkptr; i++)
1008 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1013 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1015 blkptr_t *bp = zio->io_bp;
1016 blkptr_t *bp_orig = &zio->io_bp_orig;
1019 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1020 ASSERT(BP_EQUAL(bp, bp_orig));
1022 dsl_dataset_t *ds = os->os_dsl_dataset;
1023 dmu_tx_t *tx = os->os_synctx;
1025 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1026 dsl_dataset_block_born(ds, bp, tx);
1030 /* called from dsl */
1032 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1035 zbookmark_phys_t zb;
1039 list_t *newlist = NULL;
1040 dbuf_dirty_record_t *dr;
1042 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1044 ASSERT(dmu_tx_is_syncing(tx));
1045 /* XXX the write_done callback should really give us the tx... */
1048 if (os->os_dsl_dataset == NULL) {
1050 * This is the MOS. If we have upgraded,
1051 * spa_max_replication() could change, so reset
1054 os->os_copies = spa_max_replication(os->os_spa);
1058 * Create the root block IO
1060 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1061 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1062 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1063 arc_release(os->os_phys_buf, &os->os_phys_buf);
1065 dmu_write_policy(os, NULL, 0, 0, &zp);
1067 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1068 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1069 DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready,
1070 NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE,
1071 ZIO_FLAG_MUSTSUCCEED, &zb);
1074 * Sync special dnodes - the parent IO for the sync is the root block
1076 DMU_META_DNODE(os)->dn_zio = zio;
1077 dnode_sync(DMU_META_DNODE(os), tx);
1079 os->os_phys->os_flags = os->os_flags;
1081 if (DMU_USERUSED_DNODE(os) &&
1082 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1083 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1084 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1085 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1086 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1089 txgoff = tx->tx_txg & TXG_MASK;
1091 if (dmu_objset_userused_enabled(os)) {
1092 newlist = &os->os_synced_dnodes;
1094 * We must create the list here because it uses the
1095 * dn_dirty_link[] of this txg.
1097 list_create(newlist, sizeof (dnode_t),
1098 offsetof(dnode_t, dn_dirty_link[txgoff]));
1101 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1102 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1104 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1105 while (dr = list_head(list)) {
1106 ASSERT0(dr->dr_dbuf->db_level);
1107 list_remove(list, dr);
1109 zio_nowait(dr->dr_zio);
1112 * Free intent log blocks up to this tx.
1114 zil_sync(os->os_zil, tx);
1115 os->os_phys->os_zil_header = os->os_zil_header;
1120 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1122 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1123 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1126 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1129 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1135 dmu_objset_userused_enabled(objset_t *os)
1137 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1138 used_cbs[os->os_phys->os_type] != NULL &&
1139 DMU_USERUSED_DNODE(os) != NULL);
1143 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1144 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1146 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1147 int64_t delta = DNODE_SIZE + used;
1150 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1152 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1158 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1161 list_t *list = &os->os_synced_dnodes;
1163 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1165 while (dn = list_head(list)) {
1167 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1168 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1169 dn->dn_phys->dn_flags &
1170 DNODE_FLAG_USERUSED_ACCOUNTED);
1172 /* Allocate the user/groupused objects if necessary. */
1173 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1174 VERIFY(0 == zap_create_claim(os,
1175 DMU_USERUSED_OBJECT,
1176 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1177 VERIFY(0 == zap_create_claim(os,
1178 DMU_GROUPUSED_OBJECT,
1179 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1183 * We intentionally modify the zap object even if the
1184 * net delta is zero. Otherwise
1185 * the block of the zap obj could be shared between
1186 * datasets but need to be different between them after
1190 flags = dn->dn_id_flags;
1192 if (flags & DN_ID_OLD_EXIST) {
1193 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1194 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1196 if (flags & DN_ID_NEW_EXIST) {
1197 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1198 dn->dn_phys->dn_flags, dn->dn_newuid,
1199 dn->dn_newgid, B_FALSE, tx);
1202 mutex_enter(&dn->dn_mtx);
1204 dn->dn_oldflags = 0;
1205 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1206 dn->dn_olduid = dn->dn_newuid;
1207 dn->dn_oldgid = dn->dn_newgid;
1208 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1209 if (dn->dn_bonuslen == 0)
1210 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1212 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1214 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1215 mutex_exit(&dn->dn_mtx);
1217 list_remove(list, dn);
1218 dnode_rele(dn, list);
1223 * Returns a pointer to data to find uid/gid from
1225 * If a dirty record for transaction group that is syncing can't
1226 * be found then NULL is returned. In the NULL case it is assumed
1227 * the uid/gid aren't changing.
1230 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1232 dbuf_dirty_record_t *dr, **drp;
1235 if (db->db_dirtycnt == 0)
1236 return (db->db.db_data); /* Nothing is changing */
1238 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1239 if (dr->dr_txg == tx->tx_txg)
1247 DB_DNODE_ENTER(dr->dr_dbuf);
1248 dn = DB_DNODE(dr->dr_dbuf);
1250 if (dn->dn_bonuslen == 0 &&
1251 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1252 data = dr->dt.dl.dr_data->b_data;
1254 data = dr->dt.dl.dr_data;
1256 DB_DNODE_EXIT(dr->dr_dbuf);
1263 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1265 objset_t *os = dn->dn_objset;
1267 dmu_buf_impl_t *db = NULL;
1268 uint64_t *user = NULL;
1269 uint64_t *group = NULL;
1270 int flags = dn->dn_id_flags;
1272 boolean_t have_spill = B_FALSE;
1274 if (!dmu_objset_userused_enabled(dn->dn_objset))
1277 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1278 DN_ID_CHKED_SPILL)))
1281 if (before && dn->dn_bonuslen != 0)
1282 data = DN_BONUS(dn->dn_phys);
1283 else if (!before && dn->dn_bonuslen != 0) {
1286 mutex_enter(&db->db_mtx);
1287 data = dmu_objset_userquota_find_data(db, tx);
1289 data = DN_BONUS(dn->dn_phys);
1291 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1294 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1295 rf |= DB_RF_HAVESTRUCT;
1296 error = dmu_spill_hold_by_dnode(dn,
1297 rf | DB_RF_MUST_SUCCEED,
1298 FTAG, (dmu_buf_t **)&db);
1300 mutex_enter(&db->db_mtx);
1301 data = (before) ? db->db.db_data :
1302 dmu_objset_userquota_find_data(db, tx);
1303 have_spill = B_TRUE;
1305 mutex_enter(&dn->dn_mtx);
1306 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1307 mutex_exit(&dn->dn_mtx);
1313 user = &dn->dn_olduid;
1314 group = &dn->dn_oldgid;
1316 user = &dn->dn_newuid;
1317 group = &dn->dn_newgid;
1321 * Must always call the callback in case the object
1322 * type has changed and that type isn't an object type to track
1324 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1328 * Preserve existing uid/gid when the callback can't determine
1329 * what the new uid/gid are and the callback returned EEXIST.
1330 * The EEXIST error tells us to just use the existing uid/gid.
1331 * If we don't know what the old values are then just assign
1332 * them to 0, since that is a new file being created.
1334 if (!before && data == NULL && error == EEXIST) {
1335 if (flags & DN_ID_OLD_EXIST) {
1336 dn->dn_newuid = dn->dn_olduid;
1337 dn->dn_newgid = dn->dn_oldgid;
1346 mutex_exit(&db->db_mtx);
1348 mutex_enter(&dn->dn_mtx);
1349 if (error == 0 && before)
1350 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1351 if (error == 0 && !before)
1352 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1355 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1357 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1359 mutex_exit(&dn->dn_mtx);
1361 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1365 dmu_objset_userspace_present(objset_t *os)
1367 return (os->os_phys->os_flags &
1368 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1372 dmu_objset_userspace_upgrade(objset_t *os)
1377 if (dmu_objset_userspace_present(os))
1379 if (!dmu_objset_userused_enabled(os))
1380 return (SET_ERROR(ENOTSUP));
1381 if (dmu_objset_is_snapshot(os))
1382 return (SET_ERROR(EINVAL));
1385 * We simply need to mark every object dirty, so that it will be
1386 * synced out and now accounted. If this is called
1387 * concurrently, or if we already did some work before crashing,
1388 * that's fine, since we track each object's accounted state
1392 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1397 if (issig(JUSTLOOKING) && issig(FORREAL))
1398 return (SET_ERROR(EINTR));
1400 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1403 tx = dmu_tx_create(os);
1404 dmu_tx_hold_bonus(tx, obj);
1405 objerr = dmu_tx_assign(tx, TXG_WAIT);
1410 dmu_buf_will_dirty(db, tx);
1411 dmu_buf_rele(db, FTAG);
1415 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1416 txg_wait_synced(dmu_objset_pool(os), 0);
1421 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1422 uint64_t *usedobjsp, uint64_t *availobjsp)
1424 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1425 usedobjsp, availobjsp);
1429 dmu_objset_fsid_guid(objset_t *os)
1431 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1435 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1437 stat->dds_type = os->os_phys->os_type;
1438 if (os->os_dsl_dataset)
1439 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1443 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1445 ASSERT(os->os_dsl_dataset ||
1446 os->os_phys->os_type == DMU_OST_META);
1448 if (os->os_dsl_dataset != NULL)
1449 dsl_dataset_stats(os->os_dsl_dataset, nv);
1451 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1452 os->os_phys->os_type);
1453 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1454 dmu_objset_userspace_present(os));
1458 dmu_objset_is_snapshot(objset_t *os)
1460 if (os->os_dsl_dataset != NULL)
1461 return (dsl_dataset_is_snapshot(os->os_dsl_dataset));
1467 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1468 boolean_t *conflict)
1470 dsl_dataset_t *ds = os->os_dsl_dataset;
1473 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1474 return (SET_ERROR(ENOENT));
1476 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1477 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
1478 MT_FIRST, real, maxlen, conflict));
1482 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1483 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1485 dsl_dataset_t *ds = os->os_dsl_dataset;
1486 zap_cursor_t cursor;
1487 zap_attribute_t attr;
1489 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1491 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1492 return (SET_ERROR(ENOENT));
1494 zap_cursor_init_serialized(&cursor,
1495 ds->ds_dir->dd_pool->dp_meta_objset,
1496 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
1498 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1499 zap_cursor_fini(&cursor);
1500 return (SET_ERROR(ENOENT));
1503 if (strlen(attr.za_name) + 1 > namelen) {
1504 zap_cursor_fini(&cursor);
1505 return (SET_ERROR(ENAMETOOLONG));
1508 (void) strcpy(name, attr.za_name);
1510 *idp = attr.za_first_integer;
1512 *case_conflict = attr.za_normalization_conflict;
1513 zap_cursor_advance(&cursor);
1514 *offp = zap_cursor_serialize(&cursor);
1515 zap_cursor_fini(&cursor);
1521 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1522 uint64_t *idp, uint64_t *offp)
1524 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1525 zap_cursor_t cursor;
1526 zap_attribute_t attr;
1528 /* there is no next dir on a snapshot! */
1529 if (os->os_dsl_dataset->ds_object !=
1530 dsl_dir_phys(dd)->dd_head_dataset_obj)
1531 return (SET_ERROR(ENOENT));
1533 zap_cursor_init_serialized(&cursor,
1534 dd->dd_pool->dp_meta_objset,
1535 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
1537 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1538 zap_cursor_fini(&cursor);
1539 return (SET_ERROR(ENOENT));
1542 if (strlen(attr.za_name) + 1 > namelen) {
1543 zap_cursor_fini(&cursor);
1544 return (SET_ERROR(ENAMETOOLONG));
1547 (void) strcpy(name, attr.za_name);
1549 *idp = attr.za_first_integer;
1550 zap_cursor_advance(&cursor);
1551 *offp = zap_cursor_serialize(&cursor);
1552 zap_cursor_fini(&cursor);
1558 * Find objsets under and including ddobj, call func(ds) on each.
1561 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1562 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1567 zap_attribute_t *attr;
1571 ASSERT(dsl_pool_config_held(dp));
1573 err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd);
1577 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1578 if (dd->dd_myname[0] == '$') {
1579 dsl_dir_rele(dd, FTAG);
1583 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1584 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1587 * Iterate over all children.
1589 if (flags & DS_FIND_CHILDREN) {
1590 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1591 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1592 zap_cursor_retrieve(&zc, attr) == 0;
1593 (void) zap_cursor_advance(&zc)) {
1594 ASSERT3U(attr->za_integer_length, ==,
1596 ASSERT3U(attr->za_num_integers, ==, 1);
1598 err = dmu_objset_find_dp(dp, attr->za_first_integer,
1603 zap_cursor_fini(&zc);
1606 dsl_dir_rele(dd, FTAG);
1607 kmem_free(attr, sizeof (zap_attribute_t));
1613 * Iterate over all snapshots.
1615 if (flags & DS_FIND_SNAPSHOTS) {
1617 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1622 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1623 dsl_dataset_rele(ds, FTAG);
1625 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1626 zap_cursor_retrieve(&zc, attr) == 0;
1627 (void) zap_cursor_advance(&zc)) {
1628 ASSERT3U(attr->za_integer_length, ==,
1630 ASSERT3U(attr->za_num_integers, ==, 1);
1632 err = dsl_dataset_hold_obj(dp,
1633 attr->za_first_integer, FTAG, &ds);
1636 err = func(dp, ds, arg);
1637 dsl_dataset_rele(ds, FTAG);
1641 zap_cursor_fini(&zc);
1645 dsl_dir_rele(dd, FTAG);
1646 kmem_free(attr, sizeof (zap_attribute_t));
1654 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1657 err = func(dp, ds, arg);
1658 dsl_dataset_rele(ds, FTAG);
1663 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1664 * The dp_config_rwlock must not be held when this is called, and it
1665 * will not be held when the callback is called.
1666 * Therefore this function should only be used when the pool is not changing
1667 * (e.g. in syncing context), or the callback can deal with the possible races.
1670 dmu_objset_find_impl(spa_t *spa, const char *name,
1671 int func(const char *, void *), void *arg, int flags)
1674 dsl_pool_t *dp = spa_get_dsl(spa);
1677 zap_attribute_t *attr;
1682 dsl_pool_config_enter(dp, FTAG);
1684 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1686 dsl_pool_config_exit(dp, FTAG);
1690 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1691 if (dd->dd_myname[0] == '$') {
1692 dsl_dir_rele(dd, FTAG);
1693 dsl_pool_config_exit(dp, FTAG);
1697 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1698 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1701 * Iterate over all children.
1703 if (flags & DS_FIND_CHILDREN) {
1704 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1705 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1706 zap_cursor_retrieve(&zc, attr) == 0;
1707 (void) zap_cursor_advance(&zc)) {
1708 ASSERT3U(attr->za_integer_length, ==,
1710 ASSERT3U(attr->za_num_integers, ==, 1);
1712 child = kmem_asprintf("%s/%s", name, attr->za_name);
1713 dsl_pool_config_exit(dp, FTAG);
1714 err = dmu_objset_find_impl(spa, child,
1716 dsl_pool_config_enter(dp, FTAG);
1721 zap_cursor_fini(&zc);
1724 dsl_dir_rele(dd, FTAG);
1725 dsl_pool_config_exit(dp, FTAG);
1726 kmem_free(attr, sizeof (zap_attribute_t));
1732 * Iterate over all snapshots.
1734 if (flags & DS_FIND_SNAPSHOTS) {
1735 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1740 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1741 dsl_dataset_rele(ds, FTAG);
1743 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1744 zap_cursor_retrieve(&zc, attr) == 0;
1745 (void) zap_cursor_advance(&zc)) {
1746 ASSERT3U(attr->za_integer_length, ==,
1748 ASSERT3U(attr->za_num_integers, ==, 1);
1750 child = kmem_asprintf("%s@%s",
1751 name, attr->za_name);
1752 dsl_pool_config_exit(dp, FTAG);
1753 err = func(child, arg);
1754 dsl_pool_config_enter(dp, FTAG);
1759 zap_cursor_fini(&zc);
1763 dsl_dir_rele(dd, FTAG);
1764 kmem_free(attr, sizeof (zap_attribute_t));
1765 dsl_pool_config_exit(dp, FTAG);
1770 /* Apply to self. */
1771 return (func(name, arg));
1775 * See comment above dmu_objset_find_impl().
1778 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1784 error = spa_open(name, &spa, FTAG);
1787 error = dmu_objset_find_impl(spa, name, func, arg, flags);
1788 spa_close(spa, FTAG);
1793 dmu_objset_set_user(objset_t *os, void *user_ptr)
1795 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1796 os->os_user_ptr = user_ptr;
1800 dmu_objset_get_user(objset_t *os)
1802 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1803 return (os->os_user_ptr);
1807 * Determine name of filesystem, given name of snapshot.
1808 * buf must be at least MAXNAMELEN bytes
1811 dmu_fsname(const char *snapname, char *buf)
1813 char *atp = strchr(snapname, '@');
1815 return (SET_ERROR(EINVAL));
1816 if (atp - snapname >= MAXNAMELEN)
1817 return (SET_ERROR(ENAMETOOLONG));
1818 (void) strlcpy(buf, snapname, atp - snapname + 1);