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) 2013 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
28 /* Portions Copyright 2010 Robert Milkowski */
31 #include <sys/zfs_context.h>
32 #include <sys/dmu_objset.h>
33 #include <sys/dsl_dir.h>
34 #include <sys/dsl_dataset.h>
35 #include <sys/dsl_prop.h>
36 #include <sys/dsl_pool.h>
37 #include <sys/dsl_synctask.h>
38 #include <sys/dsl_deleg.h>
39 #include <sys/dnode.h>
42 #include <sys/dmu_tx.h>
45 #include <sys/dmu_impl.h>
46 #include <sys/zfs_ioctl.h>
48 #include <sys/zfs_onexit.h>
49 #include <sys/dsl_destroy.h>
52 * Needed to close a window in dnode_move() that allows the objset to be freed
53 * before it can be safely accessed.
60 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
70 dmu_objset_spa(objset_t *os)
76 dmu_objset_zil(objset_t *os)
82 dmu_objset_pool(objset_t *os)
86 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
87 return (ds->ds_dir->dd_pool);
89 return (spa_get_dsl(os->os_spa));
93 dmu_objset_ds(objset_t *os)
95 return (os->os_dsl_dataset);
99 dmu_objset_type(objset_t *os)
101 return (os->os_phys->os_type);
105 dmu_objset_name(objset_t *os, char *buf)
107 dsl_dataset_name(os->os_dsl_dataset, buf);
111 dmu_objset_id(objset_t *os)
113 dsl_dataset_t *ds = os->os_dsl_dataset;
115 return (ds ? ds->ds_object : 0);
119 dmu_objset_syncprop(objset_t *os)
121 return (os->os_sync);
125 dmu_objset_logbias(objset_t *os)
127 return (os->os_logbias);
131 checksum_changed_cb(void *arg, uint64_t newval)
136 * Inheritance should have been done by now.
138 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
140 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
144 compression_changed_cb(void *arg, uint64_t newval)
149 * Inheritance and range checking should have been done by now.
151 ASSERT(newval != ZIO_COMPRESS_INHERIT);
153 os->os_compress = zio_compress_select(newval, ZIO_COMPRESS_ON_VALUE);
157 copies_changed_cb(void *arg, uint64_t newval)
162 * Inheritance and range checking should have been done by now.
165 ASSERT(newval <= spa_max_replication(os->os_spa));
167 os->os_copies = newval;
171 dedup_changed_cb(void *arg, uint64_t newval)
174 spa_t *spa = os->os_spa;
175 enum zio_checksum checksum;
178 * Inheritance should have been done by now.
180 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
182 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
184 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
185 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
189 primary_cache_changed_cb(void *arg, uint64_t newval)
194 * Inheritance and range checking should have been done by now.
196 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
197 newval == ZFS_CACHE_METADATA);
199 os->os_primary_cache = newval;
203 secondary_cache_changed_cb(void *arg, uint64_t newval)
208 * Inheritance and range checking should have been done by now.
210 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
211 newval == ZFS_CACHE_METADATA);
213 os->os_secondary_cache = newval;
217 sync_changed_cb(void *arg, uint64_t newval)
222 * Inheritance and range checking should have been done by now.
224 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
225 newval == ZFS_SYNC_DISABLED);
227 os->os_sync = newval;
229 zil_set_sync(os->os_zil, newval);
233 logbias_changed_cb(void *arg, uint64_t newval)
237 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
238 newval == ZFS_LOGBIAS_THROUGHPUT);
239 os->os_logbias = newval;
241 zil_set_logbias(os->os_zil, newval);
245 dmu_objset_byteswap(void *buf, size_t size)
247 objset_phys_t *osp = buf;
249 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
250 dnode_byteswap(&osp->os_meta_dnode);
251 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
252 osp->os_type = BSWAP_64(osp->os_type);
253 osp->os_flags = BSWAP_64(osp->os_flags);
254 if (size == sizeof (objset_phys_t)) {
255 dnode_byteswap(&osp->os_userused_dnode);
256 dnode_byteswap(&osp->os_groupused_dnode);
261 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
267 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
269 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
270 os->os_dsl_dataset = ds;
273 if (!BP_IS_HOLE(os->os_rootbp)) {
274 uint32_t aflags = ARC_WAIT;
276 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
277 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
279 if (DMU_OS_IS_L2CACHEABLE(os))
280 aflags |= ARC_L2CACHE;
281 if (DMU_OS_IS_L2COMPRESSIBLE(os))
282 aflags |= ARC_L2COMPRESS;
284 dprintf_bp(os->os_rootbp, "reading %s", "");
285 err = arc_read(NULL, spa, os->os_rootbp,
286 arc_getbuf_func, &os->os_phys_buf,
287 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
289 kmem_free(os, sizeof (objset_t));
290 /* convert checksum errors into IO errors */
292 err = SET_ERROR(EIO);
296 /* Increase the blocksize if we are permitted. */
297 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
298 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
299 arc_buf_t *buf = arc_buf_alloc(spa,
300 sizeof (objset_phys_t), &os->os_phys_buf,
302 bzero(buf->b_data, sizeof (objset_phys_t));
303 bcopy(os->os_phys_buf->b_data, buf->b_data,
304 arc_buf_size(os->os_phys_buf));
305 (void) arc_buf_remove_ref(os->os_phys_buf,
307 os->os_phys_buf = buf;
310 os->os_phys = os->os_phys_buf->b_data;
311 os->os_flags = os->os_phys->os_flags;
313 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
314 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
315 os->os_phys_buf = arc_buf_alloc(spa, size,
316 &os->os_phys_buf, ARC_BUFC_METADATA);
317 os->os_phys = os->os_phys_buf->b_data;
318 bzero(os->os_phys, size);
322 * Note: the changed_cb will be called once before the register
323 * func returns, thus changing the checksum/compression from the
324 * default (fletcher2/off). Snapshots don't need to know about
325 * checksum/compression/copies.
328 err = dsl_prop_register(ds,
329 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
330 primary_cache_changed_cb, os);
332 err = dsl_prop_register(ds,
333 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
334 secondary_cache_changed_cb, os);
336 if (!dsl_dataset_is_snapshot(ds)) {
338 err = dsl_prop_register(ds,
339 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
340 checksum_changed_cb, os);
343 err = dsl_prop_register(ds,
344 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
345 compression_changed_cb, os);
348 err = dsl_prop_register(ds,
349 zfs_prop_to_name(ZFS_PROP_COPIES),
350 copies_changed_cb, os);
353 err = dsl_prop_register(ds,
354 zfs_prop_to_name(ZFS_PROP_DEDUP),
355 dedup_changed_cb, os);
358 err = dsl_prop_register(ds,
359 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
360 logbias_changed_cb, os);
363 err = dsl_prop_register(ds,
364 zfs_prop_to_name(ZFS_PROP_SYNC),
365 sync_changed_cb, os);
369 VERIFY(arc_buf_remove_ref(os->os_phys_buf,
371 kmem_free(os, sizeof (objset_t));
374 } else if (ds == NULL) {
375 /* It's the meta-objset. */
376 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
377 os->os_compress = ZIO_COMPRESS_LZJB;
378 os->os_copies = spa_max_replication(spa);
379 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
380 os->os_dedup_verify = 0;
383 os->os_primary_cache = ZFS_CACHE_ALL;
384 os->os_secondary_cache = ZFS_CACHE_ALL;
387 if (ds == NULL || !dsl_dataset_is_snapshot(ds))
388 os->os_zil_header = os->os_phys->os_zil_header;
389 os->os_zil = zil_alloc(os, &os->os_zil_header);
391 for (i = 0; i < TXG_SIZE; i++) {
392 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
393 offsetof(dnode_t, dn_dirty_link[i]));
394 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
395 offsetof(dnode_t, dn_dirty_link[i]));
397 list_create(&os->os_dnodes, sizeof (dnode_t),
398 offsetof(dnode_t, dn_link));
399 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
400 offsetof(dmu_buf_impl_t, db_link));
402 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
403 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
404 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
406 DMU_META_DNODE(os) = dnode_special_open(os,
407 &os->os_phys->os_meta_dnode, DMU_META_DNODE_OBJECT,
409 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
410 DMU_USERUSED_DNODE(os) = dnode_special_open(os,
411 &os->os_phys->os_userused_dnode, DMU_USERUSED_OBJECT,
412 &os->os_userused_dnode);
413 DMU_GROUPUSED_DNODE(os) = dnode_special_open(os,
414 &os->os_phys->os_groupused_dnode, DMU_GROUPUSED_OBJECT,
415 &os->os_groupused_dnode);
419 * We should be the only thread trying to do this because we
420 * have ds_opening_lock
423 mutex_enter(&ds->ds_lock);
424 ASSERT(ds->ds_objset == NULL);
426 mutex_exit(&ds->ds_lock);
434 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
438 mutex_enter(&ds->ds_opening_lock);
439 *osp = ds->ds_objset;
441 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
442 ds, dsl_dataset_get_blkptr(ds), osp);
444 mutex_exit(&ds->ds_opening_lock);
449 * Holds the pool while the objset is held. Therefore only one objset
450 * can be held at a time.
453 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
459 err = dsl_pool_hold(name, tag, &dp);
462 err = dsl_dataset_hold(dp, name, tag, &ds);
464 dsl_pool_rele(dp, tag);
468 err = dmu_objset_from_ds(ds, osp);
470 dsl_dataset_rele(ds, tag);
471 dsl_pool_rele(dp, tag);
478 * dsl_pool must not be held when this is called.
479 * Upon successful return, there will be a longhold on the dataset,
480 * and the dsl_pool will not be held.
483 dmu_objset_own(const char *name, dmu_objset_type_t type,
484 boolean_t readonly, void *tag, objset_t **osp)
490 err = dsl_pool_hold(name, FTAG, &dp);
493 err = dsl_dataset_own(dp, name, tag, &ds);
495 dsl_pool_rele(dp, FTAG);
499 err = dmu_objset_from_ds(ds, osp);
500 dsl_pool_rele(dp, FTAG);
502 dsl_dataset_disown(ds, tag);
503 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
504 dsl_dataset_disown(ds, tag);
505 return (SET_ERROR(EINVAL));
506 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
507 dsl_dataset_disown(ds, tag);
508 return (SET_ERROR(EROFS));
514 dmu_objset_rele(objset_t *os, void *tag)
516 dsl_pool_t *dp = dmu_objset_pool(os);
517 dsl_dataset_rele(os->os_dsl_dataset, tag);
518 dsl_pool_rele(dp, tag);
522 * When we are called, os MUST refer to an objset associated with a dataset
523 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
524 * == tag. We will then release and reacquire ownership of the dataset while
525 * holding the pool config_rwlock to avoid intervening namespace or ownership
528 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
529 * release the hold on its dataset and acquire a new one on the dataset of the
530 * same name so that it can be partially torn down and reconstructed.
533 dmu_objset_refresh_ownership(objset_t *os, void *tag)
536 dsl_dataset_t *ds, *newds;
537 char name[MAXNAMELEN];
539 ds = os->os_dsl_dataset;
540 VERIFY3P(ds, !=, NULL);
541 VERIFY3P(ds->ds_owner, ==, tag);
542 VERIFY(dsl_dataset_long_held(ds));
544 dsl_dataset_name(ds, name);
545 dp = dmu_objset_pool(os);
546 dsl_pool_config_enter(dp, FTAG);
547 dmu_objset_disown(os, tag);
548 VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
549 VERIFY3P(newds, ==, os->os_dsl_dataset);
550 dsl_pool_config_exit(dp, FTAG);
554 dmu_objset_disown(objset_t *os, void *tag)
556 dsl_dataset_disown(os->os_dsl_dataset, tag);
560 dmu_objset_evict_dbufs(objset_t *os)
564 mutex_enter(&os->os_lock);
566 /* process the mdn last, since the other dnodes have holds on it */
567 list_remove(&os->os_dnodes, DMU_META_DNODE(os));
568 list_insert_tail(&os->os_dnodes, DMU_META_DNODE(os));
571 * Find the first dnode with holds. We have to do this dance
572 * because dnode_add_ref() only works if you already have a
573 * hold. If there are no holds then it has no dbufs so OK to
576 for (dn = list_head(&os->os_dnodes);
577 dn && !dnode_add_ref(dn, FTAG);
578 dn = list_next(&os->os_dnodes, dn))
582 dnode_t *next_dn = dn;
585 next_dn = list_next(&os->os_dnodes, next_dn);
586 } while (next_dn && !dnode_add_ref(next_dn, FTAG));
588 mutex_exit(&os->os_lock);
589 dnode_evict_dbufs(dn);
590 dnode_rele(dn, FTAG);
591 mutex_enter(&os->os_lock);
594 mutex_exit(&os->os_lock);
598 dmu_objset_evict(objset_t *os)
600 dsl_dataset_t *ds = os->os_dsl_dataset;
602 for (int t = 0; t < TXG_SIZE; t++)
603 ASSERT(!dmu_objset_is_dirty(os, t));
606 if (!dsl_dataset_is_snapshot(ds)) {
607 VERIFY0(dsl_prop_unregister(ds,
608 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
609 checksum_changed_cb, os));
610 VERIFY0(dsl_prop_unregister(ds,
611 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
612 compression_changed_cb, os));
613 VERIFY0(dsl_prop_unregister(ds,
614 zfs_prop_to_name(ZFS_PROP_COPIES),
615 copies_changed_cb, os));
616 VERIFY0(dsl_prop_unregister(ds,
617 zfs_prop_to_name(ZFS_PROP_DEDUP),
618 dedup_changed_cb, os));
619 VERIFY0(dsl_prop_unregister(ds,
620 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
621 logbias_changed_cb, os));
622 VERIFY0(dsl_prop_unregister(ds,
623 zfs_prop_to_name(ZFS_PROP_SYNC),
624 sync_changed_cb, os));
626 VERIFY0(dsl_prop_unregister(ds,
627 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
628 primary_cache_changed_cb, os));
629 VERIFY0(dsl_prop_unregister(ds,
630 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
631 secondary_cache_changed_cb, os));
637 dmu_objset_evict_dbufs(os);
639 dnode_special_close(&os->os_meta_dnode);
640 if (DMU_USERUSED_DNODE(os)) {
641 dnode_special_close(&os->os_userused_dnode);
642 dnode_special_close(&os->os_groupused_dnode);
644 zil_free(os->os_zil);
646 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
648 VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
651 * This is a barrier to prevent the objset from going away in
652 * dnode_move() until we can safely ensure that the objset is still in
653 * use. We consider the objset valid before the barrier and invalid
656 rw_enter(&os_lock, RW_READER);
659 mutex_destroy(&os->os_lock);
660 mutex_destroy(&os->os_obj_lock);
661 mutex_destroy(&os->os_user_ptr_lock);
662 kmem_free(os, sizeof (objset_t));
666 dmu_objset_snap_cmtime(objset_t *os)
668 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
671 /* called from dsl for meta-objset */
673 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
674 dmu_objset_type_t type, dmu_tx_t *tx)
679 ASSERT(dmu_tx_is_syncing(tx));
682 VERIFY0(dmu_objset_from_ds(ds, &os));
684 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
686 mdn = DMU_META_DNODE(os);
688 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
689 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
692 * We don't want to have to increase the meta-dnode's nlevels
693 * later, because then we could do it in quescing context while
694 * we are also accessing it in open context.
696 * This precaution is not necessary for the MOS (ds == NULL),
697 * because the MOS is only updated in syncing context.
698 * This is most fortunate: the MOS is the only objset that
699 * needs to be synced multiple times as spa_sync() iterates
700 * to convergence, so minimizing its dn_nlevels matters.
706 * Determine the number of levels necessary for the meta-dnode
707 * to contain DN_MAX_OBJECT dnodes.
709 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
710 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
711 DN_MAX_OBJECT * sizeof (dnode_phys_t))
714 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
715 mdn->dn_nlevels = levels;
718 ASSERT(type != DMU_OST_NONE);
719 ASSERT(type != DMU_OST_ANY);
720 ASSERT(type < DMU_OST_NUMTYPES);
721 os->os_phys->os_type = type;
722 if (dmu_objset_userused_enabled(os)) {
723 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
724 os->os_flags = os->os_phys->os_flags;
727 dsl_dataset_dirty(ds, tx);
732 typedef struct dmu_objset_create_arg {
733 const char *doca_name;
735 void (*doca_userfunc)(objset_t *os, void *arg,
736 cred_t *cr, dmu_tx_t *tx);
738 dmu_objset_type_t doca_type;
740 } dmu_objset_create_arg_t;
744 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
746 dmu_objset_create_arg_t *doca = arg;
747 dsl_pool_t *dp = dmu_tx_pool(tx);
752 if (strchr(doca->doca_name, '@') != NULL)
753 return (SET_ERROR(EINVAL));
755 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
759 dsl_dir_rele(pdd, FTAG);
760 return (SET_ERROR(EEXIST));
762 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
764 dsl_dir_rele(pdd, FTAG);
770 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
772 dmu_objset_create_arg_t *doca = arg;
773 dsl_pool_t *dp = dmu_tx_pool(tx);
781 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
783 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
784 doca->doca_cred, tx);
786 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
787 bp = dsl_dataset_get_blkptr(ds);
788 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
789 ds, bp, doca->doca_type, tx);
791 if (doca->doca_userfunc != NULL) {
792 doca->doca_userfunc(os, doca->doca_userarg,
793 doca->doca_cred, tx);
796 spa_history_log_internal_ds(ds, "create", tx, "");
797 dsl_dataset_rele(ds, FTAG);
798 dsl_dir_rele(pdd, FTAG);
802 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
803 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
805 dmu_objset_create_arg_t doca;
807 doca.doca_name = name;
808 doca.doca_cred = CRED();
809 doca.doca_flags = flags;
810 doca.doca_userfunc = func;
811 doca.doca_userarg = arg;
812 doca.doca_type = type;
814 return (dsl_sync_task(name,
815 dmu_objset_create_check, dmu_objset_create_sync, &doca, 5));
818 typedef struct dmu_objset_clone_arg {
819 const char *doca_clone;
820 const char *doca_origin;
822 } dmu_objset_clone_arg_t;
826 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
828 dmu_objset_clone_arg_t *doca = arg;
832 dsl_dataset_t *origin;
833 dsl_pool_t *dp = dmu_tx_pool(tx);
835 if (strchr(doca->doca_clone, '@') != NULL)
836 return (SET_ERROR(EINVAL));
838 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
842 dsl_dir_rele(pdd, FTAG);
843 return (SET_ERROR(EEXIST));
845 /* You can't clone across pools. */
846 if (pdd->dd_pool != dp) {
847 dsl_dir_rele(pdd, FTAG);
848 return (SET_ERROR(EXDEV));
850 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
853 dsl_dir_rele(pdd, FTAG);
854 return (SET_ERROR(EDQUOT));
856 dsl_dir_rele(pdd, FTAG);
858 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
862 /* You can't clone across pools. */
863 if (origin->ds_dir->dd_pool != dp) {
864 dsl_dataset_rele(origin, FTAG);
865 return (SET_ERROR(EXDEV));
868 /* You can only clone snapshots, not the head datasets. */
869 if (!dsl_dataset_is_snapshot(origin)) {
870 dsl_dataset_rele(origin, FTAG);
871 return (SET_ERROR(EINVAL));
873 dsl_dataset_rele(origin, FTAG);
879 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
881 dmu_objset_clone_arg_t *doca = arg;
882 dsl_pool_t *dp = dmu_tx_pool(tx);
885 dsl_dataset_t *origin, *ds;
887 char namebuf[MAXNAMELEN];
889 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
890 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
892 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
893 doca->doca_cred, tx);
895 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
896 dsl_dataset_name(origin, namebuf);
897 spa_history_log_internal_ds(ds, "clone", tx,
898 "origin=%s (%llu)", namebuf, origin->ds_object);
899 dsl_dataset_rele(ds, FTAG);
900 dsl_dataset_rele(origin, FTAG);
901 dsl_dir_rele(pdd, FTAG);
905 dmu_objset_clone(const char *clone, const char *origin)
907 dmu_objset_clone_arg_t doca;
909 doca.doca_clone = clone;
910 doca.doca_origin = origin;
911 doca.doca_cred = CRED();
913 return (dsl_sync_task(clone,
914 dmu_objset_clone_check, dmu_objset_clone_sync, &doca, 5));
918 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
921 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
922 nvlist_t *snaps = fnvlist_alloc();
924 fnvlist_add_boolean(snaps, longsnap);
926 err = dsl_dataset_snapshot(snaps, NULL, NULL);
932 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
936 while (dn = list_head(list)) {
937 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
938 ASSERT(dn->dn_dbuf->db_data_pending);
940 * Initialize dn_zio outside dnode_sync() because the
941 * meta-dnode needs to set it ouside dnode_sync().
943 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
946 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
947 list_remove(list, dn);
950 (void) dnode_add_ref(dn, newlist);
951 list_insert_tail(newlist, dn);
960 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
962 blkptr_t *bp = zio->io_bp;
964 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
966 ASSERT3P(bp, ==, os->os_rootbp);
967 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
968 ASSERT0(BP_GET_LEVEL(bp));
971 * Update rootbp fill count: it should be the number of objects
972 * allocated in the object set (not counting the "special"
973 * objects that are stored in the objset_phys_t -- the meta
974 * dnode and user/group accounting objects).
977 for (int i = 0; i < dnp->dn_nblkptr; i++)
978 bp->blk_fill += dnp->dn_blkptr[i].blk_fill;
983 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
985 blkptr_t *bp = zio->io_bp;
986 blkptr_t *bp_orig = &zio->io_bp_orig;
989 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
990 ASSERT(BP_EQUAL(bp, bp_orig));
992 dsl_dataset_t *ds = os->os_dsl_dataset;
993 dmu_tx_t *tx = os->os_synctx;
995 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
996 dsl_dataset_block_born(ds, bp, tx);
1000 /* called from dsl */
1002 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1009 list_t *newlist = NULL;
1010 dbuf_dirty_record_t *dr;
1012 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1014 ASSERT(dmu_tx_is_syncing(tx));
1015 /* XXX the write_done callback should really give us the tx... */
1018 if (os->os_dsl_dataset == NULL) {
1020 * This is the MOS. If we have upgraded,
1021 * spa_max_replication() could change, so reset
1024 os->os_copies = spa_max_replication(os->os_spa);
1028 * Create the root block IO
1030 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1031 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1032 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1033 arc_release(os->os_phys_buf, &os->os_phys_buf);
1035 dmu_write_policy(os, NULL, 0, 0, &zp);
1037 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1038 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1039 DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready,
1040 NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE,
1041 ZIO_FLAG_MUSTSUCCEED, &zb);
1044 * Sync special dnodes - the parent IO for the sync is the root block
1046 DMU_META_DNODE(os)->dn_zio = zio;
1047 dnode_sync(DMU_META_DNODE(os), tx);
1049 os->os_phys->os_flags = os->os_flags;
1051 if (DMU_USERUSED_DNODE(os) &&
1052 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1053 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1054 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1055 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1056 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1059 txgoff = tx->tx_txg & TXG_MASK;
1061 if (dmu_objset_userused_enabled(os)) {
1062 newlist = &os->os_synced_dnodes;
1064 * We must create the list here because it uses the
1065 * dn_dirty_link[] of this txg.
1067 list_create(newlist, sizeof (dnode_t),
1068 offsetof(dnode_t, dn_dirty_link[txgoff]));
1071 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1072 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1074 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1075 while (dr = list_head(list)) {
1076 ASSERT0(dr->dr_dbuf->db_level);
1077 list_remove(list, dr);
1079 zio_nowait(dr->dr_zio);
1082 * Free intent log blocks up to this tx.
1084 zil_sync(os->os_zil, tx);
1085 os->os_phys->os_zil_header = os->os_zil_header;
1090 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1092 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1093 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1096 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1099 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1105 dmu_objset_userused_enabled(objset_t *os)
1107 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1108 used_cbs[os->os_phys->os_type] != NULL &&
1109 DMU_USERUSED_DNODE(os) != NULL);
1113 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1114 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1116 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1117 int64_t delta = DNODE_SIZE + used;
1120 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1122 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1128 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1131 list_t *list = &os->os_synced_dnodes;
1133 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1135 while (dn = list_head(list)) {
1137 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1138 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1139 dn->dn_phys->dn_flags &
1140 DNODE_FLAG_USERUSED_ACCOUNTED);
1142 /* Allocate the user/groupused objects if necessary. */
1143 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1144 VERIFY(0 == zap_create_claim(os,
1145 DMU_USERUSED_OBJECT,
1146 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1147 VERIFY(0 == zap_create_claim(os,
1148 DMU_GROUPUSED_OBJECT,
1149 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1153 * We intentionally modify the zap object even if the
1154 * net delta is zero. Otherwise
1155 * the block of the zap obj could be shared between
1156 * datasets but need to be different between them after
1160 flags = dn->dn_id_flags;
1162 if (flags & DN_ID_OLD_EXIST) {
1163 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1164 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1166 if (flags & DN_ID_NEW_EXIST) {
1167 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1168 dn->dn_phys->dn_flags, dn->dn_newuid,
1169 dn->dn_newgid, B_FALSE, tx);
1172 mutex_enter(&dn->dn_mtx);
1174 dn->dn_oldflags = 0;
1175 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1176 dn->dn_olduid = dn->dn_newuid;
1177 dn->dn_oldgid = dn->dn_newgid;
1178 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1179 if (dn->dn_bonuslen == 0)
1180 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1182 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1184 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1185 mutex_exit(&dn->dn_mtx);
1187 list_remove(list, dn);
1188 dnode_rele(dn, list);
1193 * Returns a pointer to data to find uid/gid from
1195 * If a dirty record for transaction group that is syncing can't
1196 * be found then NULL is returned. In the NULL case it is assumed
1197 * the uid/gid aren't changing.
1200 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1202 dbuf_dirty_record_t *dr, **drp;
1205 if (db->db_dirtycnt == 0)
1206 return (db->db.db_data); /* Nothing is changing */
1208 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1209 if (dr->dr_txg == tx->tx_txg)
1217 DB_DNODE_ENTER(dr->dr_dbuf);
1218 dn = DB_DNODE(dr->dr_dbuf);
1220 if (dn->dn_bonuslen == 0 &&
1221 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1222 data = dr->dt.dl.dr_data->b_data;
1224 data = dr->dt.dl.dr_data;
1226 DB_DNODE_EXIT(dr->dr_dbuf);
1233 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1235 objset_t *os = dn->dn_objset;
1237 dmu_buf_impl_t *db = NULL;
1238 uint64_t *user = NULL;
1239 uint64_t *group = NULL;
1240 int flags = dn->dn_id_flags;
1242 boolean_t have_spill = B_FALSE;
1244 if (!dmu_objset_userused_enabled(dn->dn_objset))
1247 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1248 DN_ID_CHKED_SPILL)))
1251 if (before && dn->dn_bonuslen != 0)
1252 data = DN_BONUS(dn->dn_phys);
1253 else if (!before && dn->dn_bonuslen != 0) {
1256 mutex_enter(&db->db_mtx);
1257 data = dmu_objset_userquota_find_data(db, tx);
1259 data = DN_BONUS(dn->dn_phys);
1261 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1264 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1265 rf |= DB_RF_HAVESTRUCT;
1266 error = dmu_spill_hold_by_dnode(dn,
1267 rf | DB_RF_MUST_SUCCEED,
1268 FTAG, (dmu_buf_t **)&db);
1270 mutex_enter(&db->db_mtx);
1271 data = (before) ? db->db.db_data :
1272 dmu_objset_userquota_find_data(db, tx);
1273 have_spill = B_TRUE;
1275 mutex_enter(&dn->dn_mtx);
1276 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1277 mutex_exit(&dn->dn_mtx);
1283 user = &dn->dn_olduid;
1284 group = &dn->dn_oldgid;
1286 user = &dn->dn_newuid;
1287 group = &dn->dn_newgid;
1291 * Must always call the callback in case the object
1292 * type has changed and that type isn't an object type to track
1294 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1298 * Preserve existing uid/gid when the callback can't determine
1299 * what the new uid/gid are and the callback returned EEXIST.
1300 * The EEXIST error tells us to just use the existing uid/gid.
1301 * If we don't know what the old values are then just assign
1302 * them to 0, since that is a new file being created.
1304 if (!before && data == NULL && error == EEXIST) {
1305 if (flags & DN_ID_OLD_EXIST) {
1306 dn->dn_newuid = dn->dn_olduid;
1307 dn->dn_newgid = dn->dn_oldgid;
1316 mutex_exit(&db->db_mtx);
1318 mutex_enter(&dn->dn_mtx);
1319 if (error == 0 && before)
1320 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1321 if (error == 0 && !before)
1322 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1325 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1327 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1329 mutex_exit(&dn->dn_mtx);
1331 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1335 dmu_objset_userspace_present(objset_t *os)
1337 return (os->os_phys->os_flags &
1338 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1342 dmu_objset_userspace_upgrade(objset_t *os)
1347 if (dmu_objset_userspace_present(os))
1349 if (!dmu_objset_userused_enabled(os))
1350 return (SET_ERROR(ENOTSUP));
1351 if (dmu_objset_is_snapshot(os))
1352 return (SET_ERROR(EINVAL));
1355 * We simply need to mark every object dirty, so that it will be
1356 * synced out and now accounted. If this is called
1357 * concurrently, or if we already did some work before crashing,
1358 * that's fine, since we track each object's accounted state
1362 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1367 if (issig(JUSTLOOKING) && issig(FORREAL))
1368 return (SET_ERROR(EINTR));
1370 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1373 tx = dmu_tx_create(os);
1374 dmu_tx_hold_bonus(tx, obj);
1375 objerr = dmu_tx_assign(tx, TXG_WAIT);
1380 dmu_buf_will_dirty(db, tx);
1381 dmu_buf_rele(db, FTAG);
1385 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1386 txg_wait_synced(dmu_objset_pool(os), 0);
1391 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1392 uint64_t *usedobjsp, uint64_t *availobjsp)
1394 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1395 usedobjsp, availobjsp);
1399 dmu_objset_fsid_guid(objset_t *os)
1401 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1405 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1407 stat->dds_type = os->os_phys->os_type;
1408 if (os->os_dsl_dataset)
1409 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1413 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1415 ASSERT(os->os_dsl_dataset ||
1416 os->os_phys->os_type == DMU_OST_META);
1418 if (os->os_dsl_dataset != NULL)
1419 dsl_dataset_stats(os->os_dsl_dataset, nv);
1421 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1422 os->os_phys->os_type);
1423 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1424 dmu_objset_userspace_present(os));
1428 dmu_objset_is_snapshot(objset_t *os)
1430 if (os->os_dsl_dataset != NULL)
1431 return (dsl_dataset_is_snapshot(os->os_dsl_dataset));
1437 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1438 boolean_t *conflict)
1440 dsl_dataset_t *ds = os->os_dsl_dataset;
1443 if (ds->ds_phys->ds_snapnames_zapobj == 0)
1444 return (SET_ERROR(ENOENT));
1446 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1447 ds->ds_phys->ds_snapnames_zapobj, name, 8, 1, &ignored, MT_FIRST,
1448 real, maxlen, conflict));
1452 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1453 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1455 dsl_dataset_t *ds = os->os_dsl_dataset;
1456 zap_cursor_t cursor;
1457 zap_attribute_t attr;
1459 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1461 if (ds->ds_phys->ds_snapnames_zapobj == 0)
1462 return (SET_ERROR(ENOENT));
1464 zap_cursor_init_serialized(&cursor,
1465 ds->ds_dir->dd_pool->dp_meta_objset,
1466 ds->ds_phys->ds_snapnames_zapobj, *offp);
1468 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1469 zap_cursor_fini(&cursor);
1470 return (SET_ERROR(ENOENT));
1473 if (strlen(attr.za_name) + 1 > namelen) {
1474 zap_cursor_fini(&cursor);
1475 return (SET_ERROR(ENAMETOOLONG));
1478 (void) strcpy(name, attr.za_name);
1480 *idp = attr.za_first_integer;
1482 *case_conflict = attr.za_normalization_conflict;
1483 zap_cursor_advance(&cursor);
1484 *offp = zap_cursor_serialize(&cursor);
1485 zap_cursor_fini(&cursor);
1491 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1492 uint64_t *idp, uint64_t *offp)
1494 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1495 zap_cursor_t cursor;
1496 zap_attribute_t attr;
1498 /* there is no next dir on a snapshot! */
1499 if (os->os_dsl_dataset->ds_object !=
1500 dd->dd_phys->dd_head_dataset_obj)
1501 return (SET_ERROR(ENOENT));
1503 zap_cursor_init_serialized(&cursor,
1504 dd->dd_pool->dp_meta_objset,
1505 dd->dd_phys->dd_child_dir_zapobj, *offp);
1507 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1508 zap_cursor_fini(&cursor);
1509 return (SET_ERROR(ENOENT));
1512 if (strlen(attr.za_name) + 1 > namelen) {
1513 zap_cursor_fini(&cursor);
1514 return (SET_ERROR(ENAMETOOLONG));
1517 (void) strcpy(name, attr.za_name);
1519 *idp = attr.za_first_integer;
1520 zap_cursor_advance(&cursor);
1521 *offp = zap_cursor_serialize(&cursor);
1522 zap_cursor_fini(&cursor);
1528 * Find objsets under and including ddobj, call func(ds) on each.
1531 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1532 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1537 zap_attribute_t *attr;
1541 ASSERT(dsl_pool_config_held(dp));
1543 err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd);
1547 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1548 if (dd->dd_myname[0] == '$') {
1549 dsl_dir_rele(dd, FTAG);
1553 thisobj = dd->dd_phys->dd_head_dataset_obj;
1554 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1557 * Iterate over all children.
1559 if (flags & DS_FIND_CHILDREN) {
1560 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1561 dd->dd_phys->dd_child_dir_zapobj);
1562 zap_cursor_retrieve(&zc, attr) == 0;
1563 (void) zap_cursor_advance(&zc)) {
1564 ASSERT3U(attr->za_integer_length, ==,
1566 ASSERT3U(attr->za_num_integers, ==, 1);
1568 err = dmu_objset_find_dp(dp, attr->za_first_integer,
1573 zap_cursor_fini(&zc);
1576 dsl_dir_rele(dd, FTAG);
1577 kmem_free(attr, sizeof (zap_attribute_t));
1583 * Iterate over all snapshots.
1585 if (flags & DS_FIND_SNAPSHOTS) {
1587 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1590 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1591 dsl_dataset_rele(ds, FTAG);
1593 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1594 zap_cursor_retrieve(&zc, attr) == 0;
1595 (void) zap_cursor_advance(&zc)) {
1596 ASSERT3U(attr->za_integer_length, ==,
1598 ASSERT3U(attr->za_num_integers, ==, 1);
1600 err = dsl_dataset_hold_obj(dp,
1601 attr->za_first_integer, FTAG, &ds);
1604 err = func(dp, ds, arg);
1605 dsl_dataset_rele(ds, FTAG);
1609 zap_cursor_fini(&zc);
1613 dsl_dir_rele(dd, FTAG);
1614 kmem_free(attr, sizeof (zap_attribute_t));
1622 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1625 err = func(dp, ds, arg);
1626 dsl_dataset_rele(ds, FTAG);
1631 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1632 * The dp_config_rwlock must not be held when this is called, and it
1633 * will not be held when the callback is called.
1634 * Therefore this function should only be used when the pool is not changing
1635 * (e.g. in syncing context), or the callback can deal with the possible races.
1638 dmu_objset_find_impl(spa_t *spa, const char *name,
1639 int func(const char *, void *), void *arg, int flags)
1642 dsl_pool_t *dp = spa_get_dsl(spa);
1645 zap_attribute_t *attr;
1650 dsl_pool_config_enter(dp, FTAG);
1652 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1654 dsl_pool_config_exit(dp, FTAG);
1658 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1659 if (dd->dd_myname[0] == '$') {
1660 dsl_dir_rele(dd, FTAG);
1661 dsl_pool_config_exit(dp, FTAG);
1665 thisobj = dd->dd_phys->dd_head_dataset_obj;
1666 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1669 * Iterate over all children.
1671 if (flags & DS_FIND_CHILDREN) {
1672 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1673 dd->dd_phys->dd_child_dir_zapobj);
1674 zap_cursor_retrieve(&zc, attr) == 0;
1675 (void) zap_cursor_advance(&zc)) {
1676 ASSERT3U(attr->za_integer_length, ==,
1678 ASSERT3U(attr->za_num_integers, ==, 1);
1680 child = kmem_asprintf("%s/%s", name, attr->za_name);
1681 dsl_pool_config_exit(dp, FTAG);
1682 err = dmu_objset_find_impl(spa, child,
1684 dsl_pool_config_enter(dp, FTAG);
1689 zap_cursor_fini(&zc);
1692 dsl_dir_rele(dd, FTAG);
1693 dsl_pool_config_exit(dp, FTAG);
1694 kmem_free(attr, sizeof (zap_attribute_t));
1700 * Iterate over all snapshots.
1702 if (flags & DS_FIND_SNAPSHOTS) {
1703 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1706 uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
1707 dsl_dataset_rele(ds, FTAG);
1709 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1710 zap_cursor_retrieve(&zc, attr) == 0;
1711 (void) zap_cursor_advance(&zc)) {
1712 ASSERT3U(attr->za_integer_length, ==,
1714 ASSERT3U(attr->za_num_integers, ==, 1);
1716 child = kmem_asprintf("%s@%s",
1717 name, attr->za_name);
1718 dsl_pool_config_exit(dp, FTAG);
1719 err = func(child, arg);
1720 dsl_pool_config_enter(dp, FTAG);
1725 zap_cursor_fini(&zc);
1729 dsl_dir_rele(dd, FTAG);
1730 kmem_free(attr, sizeof (zap_attribute_t));
1731 dsl_pool_config_exit(dp, FTAG);
1736 /* Apply to self. */
1737 return (func(name, arg));
1741 * See comment above dmu_objset_find_impl().
1744 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1750 error = spa_open(name, &spa, FTAG);
1753 error = dmu_objset_find_impl(spa, name, func, arg, flags);
1754 spa_close(spa, FTAG);
1759 dmu_objset_set_user(objset_t *os, void *user_ptr)
1761 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1762 os->os_user_ptr = user_ptr;
1766 dmu_objset_get_user(objset_t *os)
1768 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1769 return (os->os_user_ptr);
1773 * Determine name of filesystem, given name of snapshot.
1774 * buf must be at least MAXNAMELEN bytes
1777 dmu_fsname(const char *snapname, char *buf)
1779 char *atp = strchr(snapname, '@');
1781 return (SET_ERROR(EINVAL));
1782 if (atp - snapname >= MAXNAMELEN)
1783 return (SET_ERROR(ENAMETOOLONG));
1784 (void) strlcpy(buf, snapname, atp - snapname + 1);