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;
71 * Backfill lower metadnode objects after this many have been freed.
72 * Backfilling negatively impacts object creation rates, so only do it
73 * if there are enough holes to fill.
75 int dmu_rescan_dnode_threshold = 131072;
77 static void dmu_objset_find_dp_cb(void *arg);
82 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
92 dmu_objset_spa(objset_t *os)
98 dmu_objset_zil(objset_t *os)
104 dmu_objset_pool(objset_t *os)
108 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
109 return (ds->ds_dir->dd_pool);
111 return (spa_get_dsl(os->os_spa));
115 dmu_objset_ds(objset_t *os)
117 return (os->os_dsl_dataset);
121 dmu_objset_type(objset_t *os)
123 return (os->os_phys->os_type);
127 dmu_objset_name(objset_t *os, char *buf)
129 dsl_dataset_name(os->os_dsl_dataset, buf);
133 dmu_objset_id(objset_t *os)
135 dsl_dataset_t *ds = os->os_dsl_dataset;
137 return (ds ? ds->ds_object : 0);
141 dmu_objset_syncprop(objset_t *os)
143 return (os->os_sync);
147 dmu_objset_logbias(objset_t *os)
149 return (os->os_logbias);
153 checksum_changed_cb(void *arg, uint64_t newval)
158 * Inheritance should have been done by now.
160 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
162 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
166 compression_changed_cb(void *arg, uint64_t newval)
171 * Inheritance and range checking should have been done by now.
173 ASSERT(newval != ZIO_COMPRESS_INHERIT);
175 os->os_compress = zio_compress_select(os->os_spa, newval,
180 copies_changed_cb(void *arg, uint64_t newval)
185 * Inheritance and range checking should have been done by now.
188 ASSERT(newval <= spa_max_replication(os->os_spa));
190 os->os_copies = newval;
194 dedup_changed_cb(void *arg, uint64_t newval)
197 spa_t *spa = os->os_spa;
198 enum zio_checksum checksum;
201 * Inheritance should have been done by now.
203 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
205 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
207 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
208 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
212 primary_cache_changed_cb(void *arg, uint64_t newval)
217 * Inheritance and range checking should have been done by now.
219 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
220 newval == ZFS_CACHE_METADATA);
222 os->os_primary_cache = newval;
226 secondary_cache_changed_cb(void *arg, uint64_t newval)
231 * Inheritance and range checking should have been done by now.
233 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
234 newval == ZFS_CACHE_METADATA);
236 os->os_secondary_cache = newval;
240 sync_changed_cb(void *arg, uint64_t newval)
245 * Inheritance and range checking should have been done by now.
247 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
248 newval == ZFS_SYNC_DISABLED);
250 os->os_sync = newval;
252 zil_set_sync(os->os_zil, newval);
256 redundant_metadata_changed_cb(void *arg, uint64_t newval)
261 * Inheritance and range checking should have been done by now.
263 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
264 newval == ZFS_REDUNDANT_METADATA_MOST);
266 os->os_redundant_metadata = newval;
270 logbias_changed_cb(void *arg, uint64_t newval)
274 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
275 newval == ZFS_LOGBIAS_THROUGHPUT);
276 os->os_logbias = newval;
278 zil_set_logbias(os->os_zil, newval);
282 recordsize_changed_cb(void *arg, uint64_t newval)
286 os->os_recordsize = newval;
290 dmu_objset_byteswap(void *buf, size_t size)
292 objset_phys_t *osp = buf;
294 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
295 dnode_byteswap(&osp->os_meta_dnode);
296 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
297 osp->os_type = BSWAP_64(osp->os_type);
298 osp->os_flags = BSWAP_64(osp->os_flags);
299 if (size == sizeof (objset_phys_t)) {
300 dnode_byteswap(&osp->os_userused_dnode);
301 dnode_byteswap(&osp->os_groupused_dnode);
306 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
312 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
314 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
315 os->os_dsl_dataset = ds;
318 if (!BP_IS_HOLE(os->os_rootbp)) {
319 arc_flags_t aflags = ARC_FLAG_WAIT;
321 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
322 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
324 if (DMU_OS_IS_L2CACHEABLE(os))
325 aflags |= ARC_FLAG_L2CACHE;
327 dprintf_bp(os->os_rootbp, "reading %s", "");
328 err = arc_read(NULL, spa, os->os_rootbp,
329 arc_getbuf_func, &os->os_phys_buf,
330 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
332 kmem_free(os, sizeof (objset_t));
333 /* convert checksum errors into IO errors */
335 err = SET_ERROR(EIO);
339 /* Increase the blocksize if we are permitted. */
340 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
341 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
342 arc_buf_t *buf = arc_alloc_buf(spa,
343 sizeof (objset_phys_t), &os->os_phys_buf,
345 bzero(buf->b_data, sizeof (objset_phys_t));
346 bcopy(os->os_phys_buf->b_data, buf->b_data,
347 arc_buf_size(os->os_phys_buf));
348 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
349 os->os_phys_buf = buf;
352 os->os_phys = os->os_phys_buf->b_data;
353 os->os_flags = os->os_phys->os_flags;
355 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
356 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
357 os->os_phys_buf = arc_alloc_buf(spa, size,
358 &os->os_phys_buf, ARC_BUFC_METADATA);
359 os->os_phys = os->os_phys_buf->b_data;
360 bzero(os->os_phys, size);
364 * Note: the changed_cb will be called once before the register
365 * func returns, thus changing the checksum/compression from the
366 * default (fletcher2/off). Snapshots don't need to know about
367 * checksum/compression/copies.
370 boolean_t needlock = B_FALSE;
373 * Note: it's valid to open the objset if the dataset is
374 * long-held, in which case the pool_config lock will not
377 if (!dsl_pool_config_held(dmu_objset_pool(os))) {
379 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
381 err = dsl_prop_register(ds,
382 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
383 primary_cache_changed_cb, os);
385 err = dsl_prop_register(ds,
386 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
387 secondary_cache_changed_cb, os);
389 if (!ds->ds_is_snapshot) {
391 err = dsl_prop_register(ds,
392 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
393 checksum_changed_cb, os);
396 err = dsl_prop_register(ds,
397 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
398 compression_changed_cb, os);
401 err = dsl_prop_register(ds,
402 zfs_prop_to_name(ZFS_PROP_COPIES),
403 copies_changed_cb, os);
406 err = dsl_prop_register(ds,
407 zfs_prop_to_name(ZFS_PROP_DEDUP),
408 dedup_changed_cb, os);
411 err = dsl_prop_register(ds,
412 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
413 logbias_changed_cb, os);
416 err = dsl_prop_register(ds,
417 zfs_prop_to_name(ZFS_PROP_SYNC),
418 sync_changed_cb, os);
421 err = dsl_prop_register(ds,
423 ZFS_PROP_REDUNDANT_METADATA),
424 redundant_metadata_changed_cb, os);
427 err = dsl_prop_register(ds,
428 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
429 recordsize_changed_cb, os);
433 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
435 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
436 kmem_free(os, sizeof (objset_t));
440 /* It's the meta-objset. */
441 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
442 os->os_compress = ZIO_COMPRESS_ON;
443 os->os_copies = spa_max_replication(spa);
444 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
445 os->os_dedup_verify = B_FALSE;
446 os->os_logbias = ZFS_LOGBIAS_LATENCY;
447 os->os_sync = ZFS_SYNC_STANDARD;
448 os->os_primary_cache = ZFS_CACHE_ALL;
449 os->os_secondary_cache = ZFS_CACHE_ALL;
452 if (ds == NULL || !ds->ds_is_snapshot)
453 os->os_zil_header = os->os_phys->os_zil_header;
454 os->os_zil = zil_alloc(os, &os->os_zil_header);
456 for (i = 0; i < TXG_SIZE; i++) {
457 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
458 offsetof(dnode_t, dn_dirty_link[i]));
459 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
460 offsetof(dnode_t, dn_dirty_link[i]));
462 list_create(&os->os_dnodes, sizeof (dnode_t),
463 offsetof(dnode_t, dn_link));
464 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
465 offsetof(dmu_buf_impl_t, db_link));
467 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
468 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
469 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
471 dnode_special_open(os, &os->os_phys->os_meta_dnode,
472 DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
473 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
474 dnode_special_open(os, &os->os_phys->os_userused_dnode,
475 DMU_USERUSED_OBJECT, &os->os_userused_dnode);
476 dnode_special_open(os, &os->os_phys->os_groupused_dnode,
477 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
485 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
490 * We shouldn't be doing anything with dsl_dataset_t's unless the
491 * pool_config lock is held, or the dataset is long-held.
493 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) ||
494 dsl_dataset_long_held(ds));
496 mutex_enter(&ds->ds_opening_lock);
497 if (ds->ds_objset == NULL) {
499 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
500 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
501 ds, dsl_dataset_get_blkptr(ds), &os);
502 rrw_exit(&ds->ds_bp_rwlock, FTAG);
505 mutex_enter(&ds->ds_lock);
506 ASSERT(ds->ds_objset == NULL);
508 mutex_exit(&ds->ds_lock);
511 *osp = ds->ds_objset;
512 mutex_exit(&ds->ds_opening_lock);
517 * Holds the pool while the objset is held. Therefore only one objset
518 * can be held at a time.
521 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
527 err = dsl_pool_hold(name, tag, &dp);
530 err = dsl_dataset_hold(dp, name, tag, &ds);
532 dsl_pool_rele(dp, tag);
536 err = dmu_objset_from_ds(ds, osp);
538 dsl_dataset_rele(ds, tag);
539 dsl_pool_rele(dp, tag);
546 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type,
547 boolean_t readonly, void *tag, objset_t **osp)
551 err = dmu_objset_from_ds(ds, osp);
553 dsl_dataset_disown(ds, tag);
554 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
555 dsl_dataset_disown(ds, tag);
556 return (SET_ERROR(EINVAL));
557 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
558 dsl_dataset_disown(ds, tag);
559 return (SET_ERROR(EROFS));
565 * dsl_pool must not be held when this is called.
566 * Upon successful return, there will be a longhold on the dataset,
567 * and the dsl_pool will not be held.
570 dmu_objset_own(const char *name, dmu_objset_type_t type,
571 boolean_t readonly, void *tag, objset_t **osp)
577 err = dsl_pool_hold(name, FTAG, &dp);
580 err = dsl_dataset_own(dp, name, tag, &ds);
582 dsl_pool_rele(dp, FTAG);
585 err = dmu_objset_own_impl(ds, type, readonly, tag, osp);
586 dsl_pool_rele(dp, FTAG);
592 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type,
593 boolean_t readonly, void *tag, objset_t **osp)
598 err = dsl_dataset_own_obj(dp, obj, tag, &ds);
602 return (dmu_objset_own_impl(ds, type, readonly, tag, osp));
606 dmu_objset_rele(objset_t *os, void *tag)
608 dsl_pool_t *dp = dmu_objset_pool(os);
609 dsl_dataset_rele(os->os_dsl_dataset, tag);
610 dsl_pool_rele(dp, tag);
614 * When we are called, os MUST refer to an objset associated with a dataset
615 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
616 * == tag. We will then release and reacquire ownership of the dataset while
617 * holding the pool config_rwlock to avoid intervening namespace or ownership
620 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
621 * release the hold on its dataset and acquire a new one on the dataset of the
622 * same name so that it can be partially torn down and reconstructed.
625 dmu_objset_refresh_ownership(objset_t *os, void *tag)
628 dsl_dataset_t *ds, *newds;
629 char name[ZFS_MAX_DATASET_NAME_LEN];
631 ds = os->os_dsl_dataset;
632 VERIFY3P(ds, !=, NULL);
633 VERIFY3P(ds->ds_owner, ==, tag);
634 VERIFY(dsl_dataset_long_held(ds));
636 dsl_dataset_name(ds, name);
637 dp = dmu_objset_pool(os);
638 dsl_pool_config_enter(dp, FTAG);
639 dmu_objset_disown(os, tag);
640 VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
641 VERIFY3P(newds, ==, os->os_dsl_dataset);
642 dsl_pool_config_exit(dp, FTAG);
646 dmu_objset_disown(objset_t *os, void *tag)
648 dsl_dataset_disown(os->os_dsl_dataset, tag);
652 dmu_objset_evict_dbufs(objset_t *os)
657 mutex_enter(&os->os_lock);
658 dn = list_head(&os->os_dnodes);
661 * Skip dnodes without holds. We have to do this dance
662 * because dnode_add_ref() only works if there is already a
663 * hold. If the dnode has no holds, then it has no dbufs.
665 if (dnode_add_ref(dn, FTAG)) {
666 list_insert_after(&os->os_dnodes, dn, &dn_marker);
667 mutex_exit(&os->os_lock);
669 dnode_evict_dbufs(dn);
670 dnode_rele(dn, FTAG);
672 mutex_enter(&os->os_lock);
673 dn = list_next(&os->os_dnodes, &dn_marker);
674 list_remove(&os->os_dnodes, &dn_marker);
676 dn = list_next(&os->os_dnodes, dn);
679 mutex_exit(&os->os_lock);
681 if (DMU_USERUSED_DNODE(os) != NULL) {
682 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
683 dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
685 dnode_evict_dbufs(DMU_META_DNODE(os));
689 * Objset eviction processing is split into into two pieces.
690 * The first marks the objset as evicting, evicts any dbufs that
691 * have a refcount of zero, and then queues up the objset for the
692 * second phase of eviction. Once os->os_dnodes has been cleared by
693 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
694 * The second phase closes the special dnodes, dequeues the objset from
695 * the list of those undergoing eviction, and finally frees the objset.
697 * NOTE: Due to asynchronous eviction processing (invocation of
698 * dnode_buf_pageout()), it is possible for the meta dnode for the
699 * objset to have no holds even though os->os_dnodes is not empty.
702 dmu_objset_evict(objset_t *os)
704 dsl_dataset_t *ds = os->os_dsl_dataset;
706 for (int t = 0; t < TXG_SIZE; t++)
707 ASSERT(!dmu_objset_is_dirty(os, t));
710 dsl_prop_unregister_all(ds, os);
715 dmu_objset_evict_dbufs(os);
717 mutex_enter(&os->os_lock);
718 spa_evicting_os_register(os->os_spa, os);
719 if (list_is_empty(&os->os_dnodes)) {
720 mutex_exit(&os->os_lock);
721 dmu_objset_evict_done(os);
723 mutex_exit(&os->os_lock);
728 dmu_objset_evict_done(objset_t *os)
730 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
732 dnode_special_close(&os->os_meta_dnode);
733 if (DMU_USERUSED_DNODE(os)) {
734 dnode_special_close(&os->os_userused_dnode);
735 dnode_special_close(&os->os_groupused_dnode);
737 zil_free(os->os_zil);
739 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
742 * This is a barrier to prevent the objset from going away in
743 * dnode_move() until we can safely ensure that the objset is still in
744 * use. We consider the objset valid before the barrier and invalid
747 rw_enter(&os_lock, RW_READER);
750 mutex_destroy(&os->os_lock);
751 mutex_destroy(&os->os_obj_lock);
752 mutex_destroy(&os->os_user_ptr_lock);
753 spa_evicting_os_deregister(os->os_spa, os);
754 kmem_free(os, sizeof (objset_t));
758 dmu_objset_snap_cmtime(objset_t *os)
760 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
763 /* called from dsl for meta-objset */
765 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
766 dmu_objset_type_t type, dmu_tx_t *tx)
771 ASSERT(dmu_tx_is_syncing(tx));
774 VERIFY0(dmu_objset_from_ds(ds, &os));
776 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
778 mdn = DMU_META_DNODE(os);
780 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
781 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
784 * We don't want to have to increase the meta-dnode's nlevels
785 * later, because then we could do it in quescing context while
786 * we are also accessing it in open context.
788 * This precaution is not necessary for the MOS (ds == NULL),
789 * because the MOS is only updated in syncing context.
790 * This is most fortunate: the MOS is the only objset that
791 * needs to be synced multiple times as spa_sync() iterates
792 * to convergence, so minimizing its dn_nlevels matters.
798 * Determine the number of levels necessary for the meta-dnode
799 * to contain DN_MAX_OBJECT dnodes. Note that in order to
800 * ensure that we do not overflow 64 bits, there has to be
801 * a nlevels that gives us a number of blocks > DN_MAX_OBJECT
802 * but < 2^64. Therefore,
803 * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) (10) must be
804 * less than (64 - log2(DN_MAX_OBJECT)) (16).
806 while ((uint64_t)mdn->dn_nblkptr <<
807 (mdn->dn_datablkshift - DNODE_SHIFT +
808 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
812 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
813 mdn->dn_nlevels = levels;
816 ASSERT(type != DMU_OST_NONE);
817 ASSERT(type != DMU_OST_ANY);
818 ASSERT(type < DMU_OST_NUMTYPES);
819 os->os_phys->os_type = type;
820 if (dmu_objset_userused_enabled(os)) {
821 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
822 os->os_flags = os->os_phys->os_flags;
825 dsl_dataset_dirty(ds, tx);
830 typedef struct dmu_objset_create_arg {
831 const char *doca_name;
833 void (*doca_userfunc)(objset_t *os, void *arg,
834 cred_t *cr, dmu_tx_t *tx);
836 dmu_objset_type_t doca_type;
838 } dmu_objset_create_arg_t;
842 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
844 dmu_objset_create_arg_t *doca = arg;
845 dsl_pool_t *dp = dmu_tx_pool(tx);
850 if (strchr(doca->doca_name, '@') != NULL)
851 return (SET_ERROR(EINVAL));
853 if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN)
854 return (SET_ERROR(ENAMETOOLONG));
856 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
860 dsl_dir_rele(pdd, FTAG);
861 return (SET_ERROR(EEXIST));
863 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
865 dsl_dir_rele(pdd, FTAG);
871 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
873 dmu_objset_create_arg_t *doca = arg;
874 dsl_pool_t *dp = dmu_tx_pool(tx);
882 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
884 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
885 doca->doca_cred, tx);
887 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
888 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
889 bp = dsl_dataset_get_blkptr(ds);
890 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
891 ds, bp, doca->doca_type, tx);
892 rrw_exit(&ds->ds_bp_rwlock, FTAG);
894 if (doca->doca_userfunc != NULL) {
895 doca->doca_userfunc(os, doca->doca_userarg,
896 doca->doca_cred, tx);
899 spa_history_log_internal_ds(ds, "create", tx, "");
900 dsl_dataset_rele(ds, FTAG);
901 dsl_dir_rele(pdd, FTAG);
905 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
906 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
908 dmu_objset_create_arg_t doca;
910 doca.doca_name = name;
911 doca.doca_cred = CRED();
912 doca.doca_flags = flags;
913 doca.doca_userfunc = func;
914 doca.doca_userarg = arg;
915 doca.doca_type = type;
917 return (dsl_sync_task(name,
918 dmu_objset_create_check, dmu_objset_create_sync, &doca,
919 5, ZFS_SPACE_CHECK_NORMAL));
922 typedef struct dmu_objset_clone_arg {
923 const char *doca_clone;
924 const char *doca_origin;
926 } dmu_objset_clone_arg_t;
930 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
932 dmu_objset_clone_arg_t *doca = arg;
936 dsl_dataset_t *origin;
937 dsl_pool_t *dp = dmu_tx_pool(tx);
939 if (strchr(doca->doca_clone, '@') != NULL)
940 return (SET_ERROR(EINVAL));
942 if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN)
943 return (SET_ERROR(ENAMETOOLONG));
945 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
949 dsl_dir_rele(pdd, FTAG);
950 return (SET_ERROR(EEXIST));
953 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
956 dsl_dir_rele(pdd, FTAG);
957 return (SET_ERROR(EDQUOT));
959 dsl_dir_rele(pdd, FTAG);
961 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
965 /* You can only clone snapshots, not the head datasets. */
966 if (!origin->ds_is_snapshot) {
967 dsl_dataset_rele(origin, FTAG);
968 return (SET_ERROR(EINVAL));
970 dsl_dataset_rele(origin, FTAG);
976 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
978 dmu_objset_clone_arg_t *doca = arg;
979 dsl_pool_t *dp = dmu_tx_pool(tx);
982 dsl_dataset_t *origin, *ds;
984 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
986 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
987 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
989 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
990 doca->doca_cred, tx);
992 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
993 dsl_dataset_name(origin, namebuf);
994 spa_history_log_internal_ds(ds, "clone", tx,
995 "origin=%s (%llu)", namebuf, origin->ds_object);
996 dsl_dataset_rele(ds, FTAG);
997 dsl_dataset_rele(origin, FTAG);
998 dsl_dir_rele(pdd, FTAG);
1002 dmu_objset_clone(const char *clone, const char *origin)
1004 dmu_objset_clone_arg_t doca;
1006 doca.doca_clone = clone;
1007 doca.doca_origin = origin;
1008 doca.doca_cred = CRED();
1010 return (dsl_sync_task(clone,
1011 dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
1012 5, ZFS_SPACE_CHECK_NORMAL));
1016 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
1019 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1020 nvlist_t *snaps = fnvlist_alloc();
1022 fnvlist_add_boolean(snaps, longsnap);
1024 err = dsl_dataset_snapshot(snaps, NULL, NULL);
1025 fnvlist_free(snaps);
1030 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
1034 while (dn = list_head(list)) {
1035 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1036 ASSERT(dn->dn_dbuf->db_data_pending);
1038 * Initialize dn_zio outside dnode_sync() because the
1039 * meta-dnode needs to set it ouside dnode_sync().
1041 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1044 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1045 list_remove(list, dn);
1048 (void) dnode_add_ref(dn, newlist);
1049 list_insert_tail(newlist, dn);
1058 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1060 blkptr_t *bp = zio->io_bp;
1062 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1064 ASSERT(!BP_IS_EMBEDDED(bp));
1065 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1066 ASSERT0(BP_GET_LEVEL(bp));
1069 * Update rootbp fill count: it should be the number of objects
1070 * allocated in the object set (not counting the "special"
1071 * objects that are stored in the objset_phys_t -- the meta
1072 * dnode and user/group accounting objects).
1075 for (int i = 0; i < dnp->dn_nblkptr; i++)
1076 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1077 if (os->os_dsl_dataset != NULL)
1078 rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG);
1079 *os->os_rootbp = *bp;
1080 if (os->os_dsl_dataset != NULL)
1081 rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG);
1086 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1088 blkptr_t *bp = zio->io_bp;
1089 blkptr_t *bp_orig = &zio->io_bp_orig;
1092 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1093 ASSERT(BP_EQUAL(bp, bp_orig));
1095 dsl_dataset_t *ds = os->os_dsl_dataset;
1096 dmu_tx_t *tx = os->os_synctx;
1098 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1099 dsl_dataset_block_born(ds, bp, tx);
1101 kmem_free(bp, sizeof (*bp));
1104 /* called from dsl */
1106 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1109 zbookmark_phys_t zb;
1113 list_t *newlist = NULL;
1114 dbuf_dirty_record_t *dr;
1115 blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP);
1116 *blkptr_copy = *os->os_rootbp;
1118 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1120 ASSERT(dmu_tx_is_syncing(tx));
1121 /* XXX the write_done callback should really give us the tx... */
1124 if (os->os_dsl_dataset == NULL) {
1126 * This is the MOS. If we have upgraded,
1127 * spa_max_replication() could change, so reset
1130 os->os_copies = spa_max_replication(os->os_spa);
1134 * Create the root block IO
1136 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1137 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1138 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1139 arc_release(os->os_phys_buf, &os->os_phys_buf);
1141 dmu_write_policy(os, NULL, 0, 0, &zp);
1143 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1144 blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1145 &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done,
1146 os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
1149 * Sync special dnodes - the parent IO for the sync is the root block
1151 DMU_META_DNODE(os)->dn_zio = zio;
1152 dnode_sync(DMU_META_DNODE(os), tx);
1154 os->os_phys->os_flags = os->os_flags;
1156 if (DMU_USERUSED_DNODE(os) &&
1157 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1158 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1159 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1160 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1161 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1164 txgoff = tx->tx_txg & TXG_MASK;
1166 if (dmu_objset_userused_enabled(os)) {
1167 newlist = &os->os_synced_dnodes;
1169 * We must create the list here because it uses the
1170 * dn_dirty_link[] of this txg.
1172 list_create(newlist, sizeof (dnode_t),
1173 offsetof(dnode_t, dn_dirty_link[txgoff]));
1176 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1177 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1179 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1180 while (dr = list_head(list)) {
1181 ASSERT0(dr->dr_dbuf->db_level);
1182 list_remove(list, dr);
1184 zio_nowait(dr->dr_zio);
1187 /* Enable dnode backfill if enough objects have been freed. */
1188 if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) {
1189 os->os_rescan_dnodes = B_TRUE;
1190 os->os_freed_dnodes = 0;
1194 * Free intent log blocks up to this tx.
1196 zil_sync(os->os_zil, tx);
1197 os->os_phys->os_zil_header = os->os_zil_header;
1202 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1204 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1205 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1208 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1211 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1217 dmu_objset_userused_enabled(objset_t *os)
1219 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1220 used_cbs[os->os_phys->os_type] != NULL &&
1221 DMU_USERUSED_DNODE(os) != NULL);
1224 typedef struct userquota_node {
1227 avl_node_t uqn_node;
1230 typedef struct userquota_cache {
1231 avl_tree_t uqc_user_deltas;
1232 avl_tree_t uqc_group_deltas;
1233 } userquota_cache_t;
1236 userquota_compare(const void *l, const void *r)
1238 const userquota_node_t *luqn = l;
1239 const userquota_node_t *ruqn = r;
1241 if (luqn->uqn_id < ruqn->uqn_id)
1243 if (luqn->uqn_id > ruqn->uqn_id)
1249 do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx)
1252 userquota_node_t *uqn;
1254 ASSERT(dmu_tx_is_syncing(tx));
1257 while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas,
1258 &cookie)) != NULL) {
1259 VERIFY0(zap_increment_int(os, DMU_USERUSED_OBJECT,
1260 uqn->uqn_id, uqn->uqn_delta, tx));
1261 kmem_free(uqn, sizeof (*uqn));
1263 avl_destroy(&cache->uqc_user_deltas);
1266 while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas,
1267 &cookie)) != NULL) {
1268 VERIFY0(zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1269 uqn->uqn_id, uqn->uqn_delta, tx));
1270 kmem_free(uqn, sizeof (*uqn));
1272 avl_destroy(&cache->uqc_group_deltas);
1276 userquota_update_cache(avl_tree_t *avl, uint64_t id, int64_t delta)
1278 userquota_node_t search = { .uqn_id = id };
1281 userquota_node_t *uqn = avl_find(avl, &search, &idx);
1283 uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP);
1285 avl_insert(avl, uqn, idx);
1287 uqn->uqn_delta += delta;
1291 do_userquota_update(userquota_cache_t *cache, uint64_t used, uint64_t flags,
1292 uint64_t user, uint64_t group, boolean_t subtract)
1294 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1295 int64_t delta = DNODE_SIZE + used;
1299 userquota_update_cache(&cache->uqc_user_deltas, user, delta);
1300 userquota_update_cache(&cache->uqc_group_deltas, group, delta);
1305 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1308 list_t *list = &os->os_synced_dnodes;
1309 userquota_cache_t cache = { 0 };
1311 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1313 avl_create(&cache.uqc_user_deltas, userquota_compare,
1314 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1315 avl_create(&cache.uqc_group_deltas, userquota_compare,
1316 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1318 while (dn = list_head(list)) {
1320 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1321 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1322 dn->dn_phys->dn_flags &
1323 DNODE_FLAG_USERUSED_ACCOUNTED);
1325 /* Allocate the user/groupused objects if necessary. */
1326 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1327 VERIFY0(zap_create_claim(os,
1328 DMU_USERUSED_OBJECT,
1329 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1330 VERIFY0(zap_create_claim(os,
1331 DMU_GROUPUSED_OBJECT,
1332 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1335 flags = dn->dn_id_flags;
1337 if (flags & DN_ID_OLD_EXIST) {
1338 do_userquota_update(&cache,
1339 dn->dn_oldused, dn->dn_oldflags,
1340 dn->dn_olduid, dn->dn_oldgid, B_TRUE);
1342 if (flags & DN_ID_NEW_EXIST) {
1343 do_userquota_update(&cache,
1344 DN_USED_BYTES(dn->dn_phys),
1345 dn->dn_phys->dn_flags, dn->dn_newuid,
1346 dn->dn_newgid, B_FALSE);
1349 mutex_enter(&dn->dn_mtx);
1351 dn->dn_oldflags = 0;
1352 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1353 dn->dn_olduid = dn->dn_newuid;
1354 dn->dn_oldgid = dn->dn_newgid;
1355 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1356 if (dn->dn_bonuslen == 0)
1357 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1359 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1361 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1362 mutex_exit(&dn->dn_mtx);
1364 list_remove(list, dn);
1365 dnode_rele(dn, list);
1367 do_userquota_cacheflush(os, &cache, tx);
1371 * Returns a pointer to data to find uid/gid from
1373 * If a dirty record for transaction group that is syncing can't
1374 * be found then NULL is returned. In the NULL case it is assumed
1375 * the uid/gid aren't changing.
1378 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1380 dbuf_dirty_record_t *dr, **drp;
1383 if (db->db_dirtycnt == 0)
1384 return (db->db.db_data); /* Nothing is changing */
1386 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1387 if (dr->dr_txg == tx->tx_txg)
1395 DB_DNODE_ENTER(dr->dr_dbuf);
1396 dn = DB_DNODE(dr->dr_dbuf);
1398 if (dn->dn_bonuslen == 0 &&
1399 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1400 data = dr->dt.dl.dr_data->b_data;
1402 data = dr->dt.dl.dr_data;
1404 DB_DNODE_EXIT(dr->dr_dbuf);
1411 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1413 objset_t *os = dn->dn_objset;
1415 dmu_buf_impl_t *db = NULL;
1416 uint64_t *user = NULL;
1417 uint64_t *group = NULL;
1418 int flags = dn->dn_id_flags;
1420 boolean_t have_spill = B_FALSE;
1422 if (!dmu_objset_userused_enabled(dn->dn_objset))
1425 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1426 DN_ID_CHKED_SPILL)))
1429 if (before && dn->dn_bonuslen != 0)
1430 data = DN_BONUS(dn->dn_phys);
1431 else if (!before && dn->dn_bonuslen != 0) {
1434 mutex_enter(&db->db_mtx);
1435 data = dmu_objset_userquota_find_data(db, tx);
1437 data = DN_BONUS(dn->dn_phys);
1439 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1442 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1443 rf |= DB_RF_HAVESTRUCT;
1444 error = dmu_spill_hold_by_dnode(dn,
1445 rf | DB_RF_MUST_SUCCEED,
1446 FTAG, (dmu_buf_t **)&db);
1448 mutex_enter(&db->db_mtx);
1449 data = (before) ? db->db.db_data :
1450 dmu_objset_userquota_find_data(db, tx);
1451 have_spill = B_TRUE;
1453 mutex_enter(&dn->dn_mtx);
1454 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1455 mutex_exit(&dn->dn_mtx);
1461 user = &dn->dn_olduid;
1462 group = &dn->dn_oldgid;
1464 user = &dn->dn_newuid;
1465 group = &dn->dn_newgid;
1469 * Must always call the callback in case the object
1470 * type has changed and that type isn't an object type to track
1472 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1476 * Preserve existing uid/gid when the callback can't determine
1477 * what the new uid/gid are and the callback returned EEXIST.
1478 * The EEXIST error tells us to just use the existing uid/gid.
1479 * If we don't know what the old values are then just assign
1480 * them to 0, since that is a new file being created.
1482 if (!before && data == NULL && error == EEXIST) {
1483 if (flags & DN_ID_OLD_EXIST) {
1484 dn->dn_newuid = dn->dn_olduid;
1485 dn->dn_newgid = dn->dn_oldgid;
1494 mutex_exit(&db->db_mtx);
1496 mutex_enter(&dn->dn_mtx);
1497 if (error == 0 && before)
1498 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1499 if (error == 0 && !before)
1500 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1503 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1505 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1507 mutex_exit(&dn->dn_mtx);
1509 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1513 dmu_objset_userspace_present(objset_t *os)
1515 return (os->os_phys->os_flags &
1516 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1520 dmu_objset_userspace_upgrade(objset_t *os)
1525 if (dmu_objset_userspace_present(os))
1527 if (!dmu_objset_userused_enabled(os))
1528 return (SET_ERROR(ENOTSUP));
1529 if (dmu_objset_is_snapshot(os))
1530 return (SET_ERROR(EINVAL));
1533 * We simply need to mark every object dirty, so that it will be
1534 * synced out and now accounted. If this is called
1535 * concurrently, or if we already did some work before crashing,
1536 * that's fine, since we track each object's accounted state
1540 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1545 if (issig(JUSTLOOKING) && issig(FORREAL))
1546 return (SET_ERROR(EINTR));
1548 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1551 tx = dmu_tx_create(os);
1552 dmu_tx_hold_bonus(tx, obj);
1553 objerr = dmu_tx_assign(tx, TXG_WAIT);
1558 dmu_buf_will_dirty(db, tx);
1559 dmu_buf_rele(db, FTAG);
1563 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1564 txg_wait_synced(dmu_objset_pool(os), 0);
1569 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1570 uint64_t *usedobjsp, uint64_t *availobjsp)
1572 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1573 usedobjsp, availobjsp);
1577 dmu_objset_fsid_guid(objset_t *os)
1579 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1583 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1585 stat->dds_type = os->os_phys->os_type;
1586 if (os->os_dsl_dataset)
1587 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1591 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1593 ASSERT(os->os_dsl_dataset ||
1594 os->os_phys->os_type == DMU_OST_META);
1596 if (os->os_dsl_dataset != NULL)
1597 dsl_dataset_stats(os->os_dsl_dataset, nv);
1599 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1600 os->os_phys->os_type);
1601 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1602 dmu_objset_userspace_present(os));
1606 dmu_objset_is_snapshot(objset_t *os)
1608 if (os->os_dsl_dataset != NULL)
1609 return (os->os_dsl_dataset->ds_is_snapshot);
1615 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1616 boolean_t *conflict)
1618 dsl_dataset_t *ds = os->os_dsl_dataset;
1621 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1622 return (SET_ERROR(ENOENT));
1624 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1625 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
1626 MT_FIRST, real, maxlen, conflict));
1630 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1631 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1633 dsl_dataset_t *ds = os->os_dsl_dataset;
1634 zap_cursor_t cursor;
1635 zap_attribute_t attr;
1637 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1639 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1640 return (SET_ERROR(ENOENT));
1642 zap_cursor_init_serialized(&cursor,
1643 ds->ds_dir->dd_pool->dp_meta_objset,
1644 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
1646 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1647 zap_cursor_fini(&cursor);
1648 return (SET_ERROR(ENOENT));
1651 if (strlen(attr.za_name) + 1 > namelen) {
1652 zap_cursor_fini(&cursor);
1653 return (SET_ERROR(ENAMETOOLONG));
1656 (void) strcpy(name, attr.za_name);
1658 *idp = attr.za_first_integer;
1660 *case_conflict = attr.za_normalization_conflict;
1661 zap_cursor_advance(&cursor);
1662 *offp = zap_cursor_serialize(&cursor);
1663 zap_cursor_fini(&cursor);
1669 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1670 uint64_t *idp, uint64_t *offp)
1672 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1673 zap_cursor_t cursor;
1674 zap_attribute_t attr;
1676 /* there is no next dir on a snapshot! */
1677 if (os->os_dsl_dataset->ds_object !=
1678 dsl_dir_phys(dd)->dd_head_dataset_obj)
1679 return (SET_ERROR(ENOENT));
1681 zap_cursor_init_serialized(&cursor,
1682 dd->dd_pool->dp_meta_objset,
1683 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
1685 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1686 zap_cursor_fini(&cursor);
1687 return (SET_ERROR(ENOENT));
1690 if (strlen(attr.za_name) + 1 > namelen) {
1691 zap_cursor_fini(&cursor);
1692 return (SET_ERROR(ENAMETOOLONG));
1695 (void) strcpy(name, attr.za_name);
1697 *idp = attr.za_first_integer;
1698 zap_cursor_advance(&cursor);
1699 *offp = zap_cursor_serialize(&cursor);
1700 zap_cursor_fini(&cursor);
1705 typedef struct dmu_objset_find_ctx {
1709 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *);
1712 kmutex_t *dc_error_lock;
1714 } dmu_objset_find_ctx_t;
1717 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp)
1719 dsl_pool_t *dp = dcp->dc_dp;
1720 dmu_objset_find_ctx_t *child_dcp;
1724 zap_attribute_t *attr;
1728 /* don't process if there already was an error */
1729 if (*dcp->dc_error != 0)
1732 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, NULL, FTAG, &dd);
1736 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1737 if (dd->dd_myname[0] == '$') {
1738 dsl_dir_rele(dd, FTAG);
1742 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1743 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1746 * Iterate over all children.
1748 if (dcp->dc_flags & DS_FIND_CHILDREN) {
1749 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1750 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1751 zap_cursor_retrieve(&zc, attr) == 0;
1752 (void) zap_cursor_advance(&zc)) {
1753 ASSERT3U(attr->za_integer_length, ==,
1755 ASSERT3U(attr->za_num_integers, ==, 1);
1757 child_dcp = kmem_alloc(sizeof (*child_dcp), KM_SLEEP);
1759 child_dcp->dc_ddobj = attr->za_first_integer;
1760 if (dcp->dc_tq != NULL)
1761 (void) taskq_dispatch(dcp->dc_tq,
1762 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP);
1764 dmu_objset_find_dp_impl(child_dcp);
1766 zap_cursor_fini(&zc);
1770 * Iterate over all snapshots.
1772 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) {
1774 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1779 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1780 dsl_dataset_rele(ds, FTAG);
1782 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1783 zap_cursor_retrieve(&zc, attr) == 0;
1784 (void) zap_cursor_advance(&zc)) {
1785 ASSERT3U(attr->za_integer_length, ==,
1787 ASSERT3U(attr->za_num_integers, ==, 1);
1789 err = dsl_dataset_hold_obj(dp,
1790 attr->za_first_integer, FTAG, &ds);
1793 err = dcp->dc_func(dp, ds, dcp->dc_arg);
1794 dsl_dataset_rele(ds, FTAG);
1798 zap_cursor_fini(&zc);
1802 dsl_dir_rele(dd, FTAG);
1803 kmem_free(attr, sizeof (zap_attribute_t));
1811 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1814 err = dcp->dc_func(dp, ds, dcp->dc_arg);
1815 dsl_dataset_rele(ds, FTAG);
1819 mutex_enter(dcp->dc_error_lock);
1820 /* only keep first error */
1821 if (*dcp->dc_error == 0)
1822 *dcp->dc_error = err;
1823 mutex_exit(dcp->dc_error_lock);
1826 kmem_free(dcp, sizeof (*dcp));
1830 dmu_objset_find_dp_cb(void *arg)
1832 dmu_objset_find_ctx_t *dcp = arg;
1833 dsl_pool_t *dp = dcp->dc_dp;
1836 * We need to get a pool_config_lock here, as there are several
1837 * asssert(pool_config_held) down the stack. Getting a lock via
1838 * dsl_pool_config_enter is risky, as it might be stalled by a
1839 * pending writer. This would deadlock, as the write lock can
1840 * only be granted when our parent thread gives up the lock.
1841 * The _prio interface gives us priority over a pending writer.
1843 dsl_pool_config_enter_prio(dp, FTAG);
1845 dmu_objset_find_dp_impl(dcp);
1847 dsl_pool_config_exit(dp, FTAG);
1851 * Find objsets under and including ddobj, call func(ds) on each.
1852 * The order for the enumeration is completely undefined.
1853 * func is called with dsl_pool_config held.
1856 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1857 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1862 dmu_objset_find_ctx_t *dcp;
1865 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL);
1866 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP);
1869 dcp->dc_ddobj = ddobj;
1870 dcp->dc_func = func;
1872 dcp->dc_flags = flags;
1873 dcp->dc_error_lock = &err_lock;
1874 dcp->dc_error = &error;
1876 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) {
1878 * In case a write lock is held we can't make use of
1879 * parallelism, as down the stack of the worker threads
1880 * the lock is asserted via dsl_pool_config_held.
1881 * In case of a read lock this is solved by getting a read
1882 * lock in each worker thread, which isn't possible in case
1883 * of a writer lock. So we fall back to the synchronous path
1885 * In the future it might be possible to get some magic into
1886 * dsl_pool_config_held in a way that it returns true for
1887 * the worker threads so that a single lock held from this
1888 * thread suffices. For now, stay single threaded.
1890 dmu_objset_find_dp_impl(dcp);
1891 mutex_destroy(&err_lock);
1896 ntasks = dmu_find_threads;
1898 ntasks = vdev_count_leaves(dp->dp_spa) * 4;
1899 tq = taskq_create("dmu_objset_find", ntasks, minclsyspri, ntasks,
1902 kmem_free(dcp, sizeof (*dcp));
1903 mutex_destroy(&err_lock);
1905 return (SET_ERROR(ENOMEM));
1909 /* dcp will be freed by task */
1910 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP);
1913 * PORTING: this code relies on the property of taskq_wait to wait
1914 * until no more tasks are queued and no more tasks are active. As
1915 * we always queue new tasks from within other tasks, task_wait
1916 * reliably waits for the full recursion to finish, even though we
1917 * enqueue new tasks after taskq_wait has been called.
1918 * On platforms other than illumos, taskq_wait may not have this
1923 mutex_destroy(&err_lock);
1929 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1930 * The dp_config_rwlock must not be held when this is called, and it
1931 * will not be held when the callback is called.
1932 * Therefore this function should only be used when the pool is not changing
1933 * (e.g. in syncing context), or the callback can deal with the possible races.
1936 dmu_objset_find_impl(spa_t *spa, const char *name,
1937 int func(const char *, void *), void *arg, int flags)
1940 dsl_pool_t *dp = spa_get_dsl(spa);
1943 zap_attribute_t *attr;
1948 dsl_pool_config_enter(dp, FTAG);
1950 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1952 dsl_pool_config_exit(dp, FTAG);
1956 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1957 if (dd->dd_myname[0] == '$') {
1958 dsl_dir_rele(dd, FTAG);
1959 dsl_pool_config_exit(dp, FTAG);
1963 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1964 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1967 * Iterate over all children.
1969 if (flags & DS_FIND_CHILDREN) {
1970 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1971 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1972 zap_cursor_retrieve(&zc, attr) == 0;
1973 (void) zap_cursor_advance(&zc)) {
1974 ASSERT3U(attr->za_integer_length, ==,
1976 ASSERT3U(attr->za_num_integers, ==, 1);
1978 child = kmem_asprintf("%s/%s", name, attr->za_name);
1979 dsl_pool_config_exit(dp, FTAG);
1980 err = dmu_objset_find_impl(spa, child,
1982 dsl_pool_config_enter(dp, FTAG);
1987 zap_cursor_fini(&zc);
1990 dsl_dir_rele(dd, FTAG);
1991 dsl_pool_config_exit(dp, FTAG);
1992 kmem_free(attr, sizeof (zap_attribute_t));
1998 * Iterate over all snapshots.
2000 if (flags & DS_FIND_SNAPSHOTS) {
2001 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2006 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2007 dsl_dataset_rele(ds, FTAG);
2009 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2010 zap_cursor_retrieve(&zc, attr) == 0;
2011 (void) zap_cursor_advance(&zc)) {
2012 ASSERT3U(attr->za_integer_length, ==,
2014 ASSERT3U(attr->za_num_integers, ==, 1);
2016 child = kmem_asprintf("%s@%s",
2017 name, attr->za_name);
2018 dsl_pool_config_exit(dp, FTAG);
2019 err = func(child, arg);
2020 dsl_pool_config_enter(dp, FTAG);
2025 zap_cursor_fini(&zc);
2029 dsl_dir_rele(dd, FTAG);
2030 kmem_free(attr, sizeof (zap_attribute_t));
2031 dsl_pool_config_exit(dp, FTAG);
2036 /* Apply to self. */
2037 return (func(name, arg));
2041 * See comment above dmu_objset_find_impl().
2044 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
2050 error = spa_open(name, &spa, FTAG);
2053 error = dmu_objset_find_impl(spa, name, func, arg, flags);
2054 spa_close(spa, FTAG);
2059 dmu_objset_set_user(objset_t *os, void *user_ptr)
2061 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2062 os->os_user_ptr = user_ptr;
2066 dmu_objset_get_user(objset_t *os)
2068 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2069 return (os->os_user_ptr);
2073 * Determine name of filesystem, given name of snapshot.
2074 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
2077 dmu_fsname(const char *snapname, char *buf)
2079 char *atp = strchr(snapname, '@');
2081 return (SET_ERROR(EINVAL));
2082 if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN)
2083 return (SET_ERROR(ENAMETOOLONG));
2084 (void) strlcpy(buf, snapname, atp - snapname + 1);