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]
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2012, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
26 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright (c) 2015, STRATO AG, Inc. All rights reserved.
29 * Copyright (c) 2014 Integros [integros.com]
30 * Copyright 2017 Nexenta Systems, Inc.
33 /* Portions Copyright 2010 Robert Milkowski */
36 #include <sys/zfs_context.h>
37 #include <sys/dmu_objset.h>
38 #include <sys/dsl_dir.h>
39 #include <sys/dsl_dataset.h>
40 #include <sys/dsl_prop.h>
41 #include <sys/dsl_pool.h>
42 #include <sys/dsl_synctask.h>
43 #include <sys/dsl_deleg.h>
44 #include <sys/dnode.h>
47 #include <sys/dmu_tx.h>
50 #include <sys/dmu_impl.h>
51 #include <sys/zfs_ioctl.h>
53 #include <sys/zfs_onexit.h>
54 #include <sys/dsl_destroy.h>
56 #include <sys/zfeature.h>
59 * Needed to close a window in dnode_move() that allows the objset to be freed
60 * before it can be safely accessed.
65 * Tunable to overwrite the maximum number of threads for the parallization
66 * of dmu_objset_find_dp, needed to speed up the import of pools with many
68 * Default is 4 times the number of leaf vdevs.
70 int dmu_find_threads = 0;
73 * Backfill lower metadnode objects after this many have been freed.
74 * Backfilling negatively impacts object creation rates, so only do it
75 * if there are enough holes to fill.
77 int dmu_rescan_dnode_threshold = 131072;
79 static void dmu_objset_find_dp_cb(void *arg);
84 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
94 dmu_objset_spa(objset_t *os)
100 dmu_objset_zil(objset_t *os)
106 dmu_objset_pool(objset_t *os)
110 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
111 return (ds->ds_dir->dd_pool);
113 return (spa_get_dsl(os->os_spa));
117 dmu_objset_ds(objset_t *os)
119 return (os->os_dsl_dataset);
123 dmu_objset_type(objset_t *os)
125 return (os->os_phys->os_type);
129 dmu_objset_name(objset_t *os, char *buf)
131 dsl_dataset_name(os->os_dsl_dataset, buf);
135 dmu_objset_id(objset_t *os)
137 dsl_dataset_t *ds = os->os_dsl_dataset;
139 return (ds ? ds->ds_object : 0);
143 dmu_objset_syncprop(objset_t *os)
145 return (os->os_sync);
149 dmu_objset_logbias(objset_t *os)
151 return (os->os_logbias);
155 checksum_changed_cb(void *arg, uint64_t newval)
160 * Inheritance should have been done by now.
162 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
164 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
168 compression_changed_cb(void *arg, uint64_t newval)
173 * Inheritance and range checking should have been done by now.
175 ASSERT(newval != ZIO_COMPRESS_INHERIT);
177 os->os_compress = zio_compress_select(os->os_spa, newval,
182 copies_changed_cb(void *arg, uint64_t newval)
187 * Inheritance and range checking should have been done by now.
190 ASSERT(newval <= spa_max_replication(os->os_spa));
192 os->os_copies = newval;
196 dedup_changed_cb(void *arg, uint64_t newval)
199 spa_t *spa = os->os_spa;
200 enum zio_checksum checksum;
203 * Inheritance should have been done by now.
205 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
207 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
209 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
210 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
214 primary_cache_changed_cb(void *arg, uint64_t newval)
219 * Inheritance and range checking should have been done by now.
221 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
222 newval == ZFS_CACHE_METADATA);
224 os->os_primary_cache = newval;
228 secondary_cache_changed_cb(void *arg, uint64_t newval)
233 * Inheritance and range checking should have been done by now.
235 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
236 newval == ZFS_CACHE_METADATA);
238 os->os_secondary_cache = newval;
242 sync_changed_cb(void *arg, uint64_t newval)
247 * Inheritance and range checking should have been done by now.
249 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
250 newval == ZFS_SYNC_DISABLED);
252 os->os_sync = newval;
254 zil_set_sync(os->os_zil, newval);
258 redundant_metadata_changed_cb(void *arg, uint64_t newval)
263 * Inheritance and range checking should have been done by now.
265 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
266 newval == ZFS_REDUNDANT_METADATA_MOST);
268 os->os_redundant_metadata = newval;
272 logbias_changed_cb(void *arg, uint64_t newval)
276 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
277 newval == ZFS_LOGBIAS_THROUGHPUT);
278 os->os_logbias = newval;
280 zil_set_logbias(os->os_zil, newval);
284 recordsize_changed_cb(void *arg, uint64_t newval)
288 os->os_recordsize = newval;
292 dmu_objset_byteswap(void *buf, size_t size)
294 objset_phys_t *osp = buf;
296 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
297 dnode_byteswap(&osp->os_meta_dnode);
298 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
299 osp->os_type = BSWAP_64(osp->os_type);
300 osp->os_flags = BSWAP_64(osp->os_flags);
301 if (size == sizeof (objset_phys_t)) {
302 dnode_byteswap(&osp->os_userused_dnode);
303 dnode_byteswap(&osp->os_groupused_dnode);
308 * The hash is a CRC-based hash of the objset_t pointer and the object number.
311 dnode_hash(const objset_t *os, uint64_t obj)
313 uintptr_t osv = (uintptr_t)os;
314 uint64_t crc = -1ULL;
316 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
318 * The low 6 bits of the pointer don't have much entropy, because
319 * the objset_t is larger than 2^6 bytes long.
321 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF];
322 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF];
323 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF];
324 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF];
326 crc ^= (osv>>14) ^ (obj>>24);
332 dnode_multilist_index_func(multilist_t *ml, void *obj)
335 return (dnode_hash(dn->dn_objset, dn->dn_object) %
336 multilist_get_num_sublists(ml));
340 * Instantiates the objset_t in-memory structure corresponding to the
341 * objset_phys_t that's pointed to by the specified blkptr_t.
344 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
350 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
354 * The $ORIGIN dataset (if it exists) doesn't have an associated
355 * objset, so there's no reason to open it. The $ORIGIN dataset
356 * will not exist on pools older than SPA_VERSION_ORIGIN.
358 if (ds != NULL && spa_get_dsl(spa) != NULL &&
359 spa_get_dsl(spa)->dp_origin_snap != NULL) {
360 ASSERT3P(ds->ds_dir, !=,
361 spa_get_dsl(spa)->dp_origin_snap->ds_dir);
365 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
366 os->os_dsl_dataset = ds;
369 if (!BP_IS_HOLE(os->os_rootbp)) {
370 arc_flags_t aflags = ARC_FLAG_WAIT;
372 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
373 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
375 if (DMU_OS_IS_L2CACHEABLE(os))
376 aflags |= ARC_FLAG_L2CACHE;
378 dprintf_bp(os->os_rootbp, "reading %s", "");
379 err = arc_read(NULL, spa, os->os_rootbp,
380 arc_getbuf_func, &os->os_phys_buf,
381 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
383 kmem_free(os, sizeof (objset_t));
384 /* convert checksum errors into IO errors */
386 err = SET_ERROR(EIO);
390 /* Increase the blocksize if we are permitted. */
391 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
392 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
393 arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf,
394 ARC_BUFC_METADATA, sizeof (objset_phys_t));
395 bzero(buf->b_data, sizeof (objset_phys_t));
396 bcopy(os->os_phys_buf->b_data, buf->b_data,
397 arc_buf_size(os->os_phys_buf));
398 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
399 os->os_phys_buf = buf;
402 os->os_phys = os->os_phys_buf->b_data;
403 os->os_flags = os->os_phys->os_flags;
405 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
406 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
407 os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf,
408 ARC_BUFC_METADATA, size);
409 os->os_phys = os->os_phys_buf->b_data;
410 bzero(os->os_phys, size);
414 * Note: the changed_cb will be called once before the register
415 * func returns, thus changing the checksum/compression from the
416 * default (fletcher2/off). Snapshots don't need to know about
417 * checksum/compression/copies.
420 boolean_t needlock = B_FALSE;
423 * Note: it's valid to open the objset if the dataset is
424 * long-held, in which case the pool_config lock will not
427 if (!dsl_pool_config_held(dmu_objset_pool(os))) {
429 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
431 err = dsl_prop_register(ds,
432 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
433 primary_cache_changed_cb, os);
435 err = dsl_prop_register(ds,
436 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
437 secondary_cache_changed_cb, os);
439 if (!ds->ds_is_snapshot) {
441 err = dsl_prop_register(ds,
442 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
443 checksum_changed_cb, os);
446 err = dsl_prop_register(ds,
447 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
448 compression_changed_cb, os);
451 err = dsl_prop_register(ds,
452 zfs_prop_to_name(ZFS_PROP_COPIES),
453 copies_changed_cb, os);
456 err = dsl_prop_register(ds,
457 zfs_prop_to_name(ZFS_PROP_DEDUP),
458 dedup_changed_cb, os);
461 err = dsl_prop_register(ds,
462 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
463 logbias_changed_cb, os);
466 err = dsl_prop_register(ds,
467 zfs_prop_to_name(ZFS_PROP_SYNC),
468 sync_changed_cb, os);
471 err = dsl_prop_register(ds,
473 ZFS_PROP_REDUNDANT_METADATA),
474 redundant_metadata_changed_cb, os);
477 err = dsl_prop_register(ds,
478 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
479 recordsize_changed_cb, os);
483 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
485 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
486 kmem_free(os, sizeof (objset_t));
490 /* It's the meta-objset. */
491 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
492 os->os_compress = ZIO_COMPRESS_ON;
493 os->os_copies = spa_max_replication(spa);
494 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
495 os->os_dedup_verify = B_FALSE;
496 os->os_logbias = ZFS_LOGBIAS_LATENCY;
497 os->os_sync = ZFS_SYNC_STANDARD;
498 os->os_primary_cache = ZFS_CACHE_ALL;
499 os->os_secondary_cache = ZFS_CACHE_ALL;
502 * These properties will be filled in by the logic in zfs_get_zplprop()
503 * when they are queried for the first time.
505 os->os_version = OBJSET_PROP_UNINITIALIZED;
506 os->os_normalization = OBJSET_PROP_UNINITIALIZED;
507 os->os_utf8only = OBJSET_PROP_UNINITIALIZED;
508 os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED;
510 if (ds == NULL || !ds->ds_is_snapshot)
511 os->os_zil_header = os->os_phys->os_zil_header;
512 os->os_zil = zil_alloc(os, &os->os_zil_header);
514 for (i = 0; i < TXG_SIZE; i++) {
515 os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t),
516 offsetof(dnode_t, dn_dirty_link[i]),
517 dnode_multilist_index_func);
519 list_create(&os->os_dnodes, sizeof (dnode_t),
520 offsetof(dnode_t, dn_link));
521 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
522 offsetof(dmu_buf_impl_t, db_link));
524 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
525 mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL);
526 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
527 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
529 dnode_special_open(os, &os->os_phys->os_meta_dnode,
530 DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
531 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
532 dnode_special_open(os, &os->os_phys->os_userused_dnode,
533 DMU_USERUSED_OBJECT, &os->os_userused_dnode);
534 dnode_special_open(os, &os->os_phys->os_groupused_dnode,
535 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
543 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
548 * We shouldn't be doing anything with dsl_dataset_t's unless the
549 * pool_config lock is held, or the dataset is long-held.
551 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) ||
552 dsl_dataset_long_held(ds));
554 mutex_enter(&ds->ds_opening_lock);
555 if (ds->ds_objset == NULL) {
557 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
558 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
559 ds, dsl_dataset_get_blkptr(ds), &os);
560 rrw_exit(&ds->ds_bp_rwlock, FTAG);
563 mutex_enter(&ds->ds_lock);
564 ASSERT(ds->ds_objset == NULL);
566 mutex_exit(&ds->ds_lock);
569 *osp = ds->ds_objset;
570 mutex_exit(&ds->ds_opening_lock);
575 * Holds the pool while the objset is held. Therefore only one objset
576 * can be held at a time.
579 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
585 err = dsl_pool_hold(name, tag, &dp);
588 err = dsl_dataset_hold(dp, name, tag, &ds);
590 dsl_pool_rele(dp, tag);
594 err = dmu_objset_from_ds(ds, osp);
596 dsl_dataset_rele(ds, tag);
597 dsl_pool_rele(dp, tag);
604 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type,
605 boolean_t readonly, void *tag, objset_t **osp)
609 err = dmu_objset_from_ds(ds, osp);
611 dsl_dataset_disown(ds, tag);
612 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
613 dsl_dataset_disown(ds, tag);
614 return (SET_ERROR(EINVAL));
615 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
616 dsl_dataset_disown(ds, tag);
617 return (SET_ERROR(EROFS));
623 * dsl_pool must not be held when this is called.
624 * Upon successful return, there will be a longhold on the dataset,
625 * and the dsl_pool will not be held.
628 dmu_objset_own(const char *name, dmu_objset_type_t type,
629 boolean_t readonly, void *tag, objset_t **osp)
635 err = dsl_pool_hold(name, FTAG, &dp);
638 err = dsl_dataset_own(dp, name, tag, &ds);
640 dsl_pool_rele(dp, FTAG);
643 err = dmu_objset_own_impl(ds, type, readonly, tag, osp);
644 dsl_pool_rele(dp, FTAG);
650 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type,
651 boolean_t readonly, void *tag, objset_t **osp)
656 err = dsl_dataset_own_obj(dp, obj, tag, &ds);
660 return (dmu_objset_own_impl(ds, type, readonly, tag, osp));
664 dmu_objset_rele(objset_t *os, void *tag)
666 dsl_pool_t *dp = dmu_objset_pool(os);
667 dsl_dataset_rele(os->os_dsl_dataset, tag);
668 dsl_pool_rele(dp, tag);
672 * When we are called, os MUST refer to an objset associated with a dataset
673 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
674 * == tag. We will then release and reacquire ownership of the dataset while
675 * holding the pool config_rwlock to avoid intervening namespace or ownership
678 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
679 * release the hold on its dataset and acquire a new one on the dataset of the
680 * same name so that it can be partially torn down and reconstructed.
683 dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds,
687 char name[ZFS_MAX_DATASET_NAME_LEN];
689 VERIFY3P(ds, !=, NULL);
690 VERIFY3P(ds->ds_owner, ==, tag);
691 VERIFY(dsl_dataset_long_held(ds));
693 dsl_dataset_name(ds, name);
694 dp = ds->ds_dir->dd_pool;
695 dsl_pool_config_enter(dp, FTAG);
696 dsl_dataset_disown(ds, tag);
697 VERIFY0(dsl_dataset_own(dp, name, tag, newds));
698 dsl_pool_config_exit(dp, FTAG);
702 dmu_objset_disown(objset_t *os, void *tag)
704 dsl_dataset_disown(os->os_dsl_dataset, tag);
708 dmu_objset_evict_dbufs(objset_t *os)
713 mutex_enter(&os->os_lock);
714 dn = list_head(&os->os_dnodes);
717 * Skip dnodes without holds. We have to do this dance
718 * because dnode_add_ref() only works if there is already a
719 * hold. If the dnode has no holds, then it has no dbufs.
721 if (dnode_add_ref(dn, FTAG)) {
722 list_insert_after(&os->os_dnodes, dn, &dn_marker);
723 mutex_exit(&os->os_lock);
725 dnode_evict_dbufs(dn);
726 dnode_rele(dn, FTAG);
728 mutex_enter(&os->os_lock);
729 dn = list_next(&os->os_dnodes, &dn_marker);
730 list_remove(&os->os_dnodes, &dn_marker);
732 dn = list_next(&os->os_dnodes, dn);
735 mutex_exit(&os->os_lock);
737 if (DMU_USERUSED_DNODE(os) != NULL) {
738 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
739 dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
741 dnode_evict_dbufs(DMU_META_DNODE(os));
745 * Objset eviction processing is split into into two pieces.
746 * The first marks the objset as evicting, evicts any dbufs that
747 * have a refcount of zero, and then queues up the objset for the
748 * second phase of eviction. Once os->os_dnodes has been cleared by
749 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
750 * The second phase closes the special dnodes, dequeues the objset from
751 * the list of those undergoing eviction, and finally frees the objset.
753 * NOTE: Due to asynchronous eviction processing (invocation of
754 * dnode_buf_pageout()), it is possible for the meta dnode for the
755 * objset to have no holds even though os->os_dnodes is not empty.
758 dmu_objset_evict(objset_t *os)
760 dsl_dataset_t *ds = os->os_dsl_dataset;
762 for (int t = 0; t < TXG_SIZE; t++)
763 ASSERT(!dmu_objset_is_dirty(os, t));
766 dsl_prop_unregister_all(ds, os);
771 dmu_objset_evict_dbufs(os);
773 mutex_enter(&os->os_lock);
774 spa_evicting_os_register(os->os_spa, os);
775 if (list_is_empty(&os->os_dnodes)) {
776 mutex_exit(&os->os_lock);
777 dmu_objset_evict_done(os);
779 mutex_exit(&os->os_lock);
784 dmu_objset_evict_done(objset_t *os)
786 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
788 dnode_special_close(&os->os_meta_dnode);
789 if (DMU_USERUSED_DNODE(os)) {
790 dnode_special_close(&os->os_userused_dnode);
791 dnode_special_close(&os->os_groupused_dnode);
793 zil_free(os->os_zil);
795 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
798 * This is a barrier to prevent the objset from going away in
799 * dnode_move() until we can safely ensure that the objset is still in
800 * use. We consider the objset valid before the barrier and invalid
803 rw_enter(&os_lock, RW_READER);
806 mutex_destroy(&os->os_lock);
807 mutex_destroy(&os->os_userused_lock);
808 mutex_destroy(&os->os_obj_lock);
809 mutex_destroy(&os->os_user_ptr_lock);
810 for (int i = 0; i < TXG_SIZE; i++) {
811 multilist_destroy(os->os_dirty_dnodes[i]);
813 spa_evicting_os_deregister(os->os_spa, os);
814 kmem_free(os, sizeof (objset_t));
818 dmu_objset_snap_cmtime(objset_t *os)
820 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
823 /* called from dsl for meta-objset */
825 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
826 dmu_objset_type_t type, dmu_tx_t *tx)
831 ASSERT(dmu_tx_is_syncing(tx));
834 VERIFY0(dmu_objset_from_ds(ds, &os));
836 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
838 mdn = DMU_META_DNODE(os);
840 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
841 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
844 * We don't want to have to increase the meta-dnode's nlevels
845 * later, because then we could do it in quescing context while
846 * we are also accessing it in open context.
848 * This precaution is not necessary for the MOS (ds == NULL),
849 * because the MOS is only updated in syncing context.
850 * This is most fortunate: the MOS is the only objset that
851 * needs to be synced multiple times as spa_sync() iterates
852 * to convergence, so minimizing its dn_nlevels matters.
858 * Determine the number of levels necessary for the meta-dnode
859 * to contain DN_MAX_OBJECT dnodes. Note that in order to
860 * ensure that we do not overflow 64 bits, there has to be
861 * a nlevels that gives us a number of blocks > DN_MAX_OBJECT
862 * but < 2^64. Therefore,
863 * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) (10) must be
864 * less than (64 - log2(DN_MAX_OBJECT)) (16).
866 while ((uint64_t)mdn->dn_nblkptr <<
867 (mdn->dn_datablkshift - DNODE_SHIFT +
868 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
872 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
873 mdn->dn_nlevels = levels;
876 ASSERT(type != DMU_OST_NONE);
877 ASSERT(type != DMU_OST_ANY);
878 ASSERT(type < DMU_OST_NUMTYPES);
879 os->os_phys->os_type = type;
880 if (dmu_objset_userused_enabled(os)) {
881 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
882 os->os_flags = os->os_phys->os_flags;
885 dsl_dataset_dirty(ds, tx);
890 typedef struct dmu_objset_create_arg {
891 const char *doca_name;
893 void (*doca_userfunc)(objset_t *os, void *arg,
894 cred_t *cr, dmu_tx_t *tx);
896 dmu_objset_type_t doca_type;
898 } dmu_objset_create_arg_t;
902 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
904 dmu_objset_create_arg_t *doca = arg;
905 dsl_pool_t *dp = dmu_tx_pool(tx);
910 if (strchr(doca->doca_name, '@') != NULL)
911 return (SET_ERROR(EINVAL));
913 if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN)
914 return (SET_ERROR(ENAMETOOLONG));
916 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
920 dsl_dir_rele(pdd, FTAG);
921 return (SET_ERROR(EEXIST));
923 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
925 dsl_dir_rele(pdd, FTAG);
931 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
933 dmu_objset_create_arg_t *doca = arg;
934 dsl_pool_t *dp = dmu_tx_pool(tx);
942 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
944 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
945 doca->doca_cred, tx);
947 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
948 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
949 bp = dsl_dataset_get_blkptr(ds);
950 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
951 ds, bp, doca->doca_type, tx);
952 rrw_exit(&ds->ds_bp_rwlock, FTAG);
954 if (doca->doca_userfunc != NULL) {
955 doca->doca_userfunc(os, doca->doca_userarg,
956 doca->doca_cred, tx);
959 spa_history_log_internal_ds(ds, "create", tx, "");
960 dsl_dataset_rele(ds, FTAG);
961 dsl_dir_rele(pdd, FTAG);
965 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
966 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
968 dmu_objset_create_arg_t doca;
970 doca.doca_name = name;
971 doca.doca_cred = CRED();
972 doca.doca_flags = flags;
973 doca.doca_userfunc = func;
974 doca.doca_userarg = arg;
975 doca.doca_type = type;
977 return (dsl_sync_task(name,
978 dmu_objset_create_check, dmu_objset_create_sync, &doca,
979 5, ZFS_SPACE_CHECK_NORMAL));
982 typedef struct dmu_objset_clone_arg {
983 const char *doca_clone;
984 const char *doca_origin;
986 } dmu_objset_clone_arg_t;
990 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
992 dmu_objset_clone_arg_t *doca = arg;
996 dsl_dataset_t *origin;
997 dsl_pool_t *dp = dmu_tx_pool(tx);
999 if (strchr(doca->doca_clone, '@') != NULL)
1000 return (SET_ERROR(EINVAL));
1002 if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN)
1003 return (SET_ERROR(ENAMETOOLONG));
1005 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
1009 dsl_dir_rele(pdd, FTAG);
1010 return (SET_ERROR(EEXIST));
1013 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
1016 dsl_dir_rele(pdd, FTAG);
1017 return (SET_ERROR(EDQUOT));
1019 dsl_dir_rele(pdd, FTAG);
1021 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
1025 /* You can only clone snapshots, not the head datasets. */
1026 if (!origin->ds_is_snapshot) {
1027 dsl_dataset_rele(origin, FTAG);
1028 return (SET_ERROR(EINVAL));
1030 dsl_dataset_rele(origin, FTAG);
1036 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
1038 dmu_objset_clone_arg_t *doca = arg;
1039 dsl_pool_t *dp = dmu_tx_pool(tx);
1042 dsl_dataset_t *origin, *ds;
1044 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1046 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
1047 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
1049 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
1050 doca->doca_cred, tx);
1052 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
1053 dsl_dataset_name(origin, namebuf);
1054 spa_history_log_internal_ds(ds, "clone", tx,
1055 "origin=%s (%llu)", namebuf, origin->ds_object);
1056 dsl_dataset_rele(ds, FTAG);
1057 dsl_dataset_rele(origin, FTAG);
1058 dsl_dir_rele(pdd, FTAG);
1062 dmu_objset_clone(const char *clone, const char *origin)
1064 dmu_objset_clone_arg_t doca;
1066 doca.doca_clone = clone;
1067 doca.doca_origin = origin;
1068 doca.doca_cred = CRED();
1070 return (dsl_sync_task(clone,
1071 dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
1072 5, ZFS_SPACE_CHECK_NORMAL));
1076 dmu_objset_remap_indirects_impl(objset_t *os, uint64_t last_removed_txg)
1079 uint64_t object = 0;
1080 while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) {
1081 error = dmu_object_remap_indirects(os, object,
1084 * If the ZPL removed the object before we managed to dnode_hold
1085 * it, we would get an ENOENT. If the ZPL declares its intent
1086 * to remove the object (dnode_free) before we manage to
1087 * dnode_hold it, we would get an EEXIST. In either case, we
1088 * want to continue remapping the other objects in the objset;
1089 * in all other cases, we want to break early.
1091 if (error != 0 && error != ENOENT && error != EEXIST) {
1095 if (error == ESRCH) {
1102 dmu_objset_remap_indirects(const char *fsname)
1105 objset_t *os = NULL;
1106 uint64_t last_removed_txg;
1107 uint64_t remap_start_txg;
1110 error = dmu_objset_hold(fsname, FTAG, &os);
1114 dd = dmu_objset_ds(os)->ds_dir;
1116 if (!spa_feature_is_enabled(dmu_objset_spa(os),
1117 SPA_FEATURE_OBSOLETE_COUNTS)) {
1118 dmu_objset_rele(os, FTAG);
1119 return (SET_ERROR(ENOTSUP));
1122 if (dsl_dataset_is_snapshot(dmu_objset_ds(os))) {
1123 dmu_objset_rele(os, FTAG);
1124 return (SET_ERROR(EINVAL));
1128 * If there has not been a removal, we're done.
1130 last_removed_txg = spa_get_last_removal_txg(dmu_objset_spa(os));
1131 if (last_removed_txg == -1ULL) {
1132 dmu_objset_rele(os, FTAG);
1137 * If we have remapped since the last removal, we're done.
1139 if (dsl_dir_is_zapified(dd)) {
1140 uint64_t last_remap_txg;
1141 if (zap_lookup(spa_meta_objset(dmu_objset_spa(os)),
1142 dd->dd_object, DD_FIELD_LAST_REMAP_TXG,
1143 sizeof (last_remap_txg), 1, &last_remap_txg) == 0 &&
1144 last_remap_txg > last_removed_txg) {
1145 dmu_objset_rele(os, FTAG);
1150 dsl_dataset_long_hold(dmu_objset_ds(os), FTAG);
1151 dsl_pool_rele(dmu_objset_pool(os), FTAG);
1153 remap_start_txg = spa_last_synced_txg(dmu_objset_spa(os));
1154 error = dmu_objset_remap_indirects_impl(os, last_removed_txg);
1157 * We update the last_remap_txg to be the start txg so that
1158 * we can guarantee that every block older than last_remap_txg
1159 * that can be remapped has been remapped.
1161 error = dsl_dir_update_last_remap_txg(dd, remap_start_txg);
1164 dsl_dataset_long_rele(dmu_objset_ds(os), FTAG);
1165 dsl_dataset_rele(dmu_objset_ds(os), FTAG);
1171 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
1174 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1175 nvlist_t *snaps = fnvlist_alloc();
1177 fnvlist_add_boolean(snaps, longsnap);
1179 err = dsl_dataset_snapshot(snaps, NULL, NULL);
1180 fnvlist_free(snaps);
1185 dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx)
1189 while ((dn = multilist_sublist_head(list)) != NULL) {
1190 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1191 ASSERT(dn->dn_dbuf->db_data_pending);
1193 * Initialize dn_zio outside dnode_sync() because the
1194 * meta-dnode needs to set it ouside dnode_sync().
1196 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1199 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1200 multilist_sublist_remove(list, dn);
1202 multilist_t *newlist = dn->dn_objset->os_synced_dnodes;
1203 if (newlist != NULL) {
1204 (void) dnode_add_ref(dn, newlist);
1205 multilist_insert(newlist, dn);
1214 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1216 blkptr_t *bp = zio->io_bp;
1218 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1220 ASSERT(!BP_IS_EMBEDDED(bp));
1221 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1222 ASSERT0(BP_GET_LEVEL(bp));
1225 * Update rootbp fill count: it should be the number of objects
1226 * allocated in the object set (not counting the "special"
1227 * objects that are stored in the objset_phys_t -- the meta
1228 * dnode and user/group accounting objects).
1231 for (int i = 0; i < dnp->dn_nblkptr; i++)
1232 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1233 if (os->os_dsl_dataset != NULL)
1234 rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG);
1235 *os->os_rootbp = *bp;
1236 if (os->os_dsl_dataset != NULL)
1237 rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG);
1242 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1244 blkptr_t *bp = zio->io_bp;
1245 blkptr_t *bp_orig = &zio->io_bp_orig;
1248 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1249 ASSERT(BP_EQUAL(bp, bp_orig));
1251 dsl_dataset_t *ds = os->os_dsl_dataset;
1252 dmu_tx_t *tx = os->os_synctx;
1254 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1255 dsl_dataset_block_born(ds, bp, tx);
1257 kmem_free(bp, sizeof (*bp));
1260 typedef struct sync_dnodes_arg {
1261 multilist_t *sda_list;
1262 int sda_sublist_idx;
1263 multilist_t *sda_newlist;
1265 } sync_dnodes_arg_t;
1268 sync_dnodes_task(void *arg)
1270 sync_dnodes_arg_t *sda = arg;
1272 multilist_sublist_t *ms =
1273 multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx);
1275 dmu_objset_sync_dnodes(ms, sda->sda_tx);
1277 multilist_sublist_unlock(ms);
1279 kmem_free(sda, sizeof (*sda));
1283 /* called from dsl */
1285 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1288 zbookmark_phys_t zb;
1292 dbuf_dirty_record_t *dr;
1293 blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP);
1294 *blkptr_copy = *os->os_rootbp;
1296 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1298 ASSERT(dmu_tx_is_syncing(tx));
1299 /* XXX the write_done callback should really give us the tx... */
1302 if (os->os_dsl_dataset == NULL) {
1304 * This is the MOS. If we have upgraded,
1305 * spa_max_replication() could change, so reset
1308 os->os_copies = spa_max_replication(os->os_spa);
1312 * Create the root block IO
1314 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1315 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1316 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1317 arc_release(os->os_phys_buf, &os->os_phys_buf);
1319 dmu_write_policy(os, NULL, 0, 0, &zp);
1321 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1322 blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1323 &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done,
1324 os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
1327 * Sync special dnodes - the parent IO for the sync is the root block
1329 DMU_META_DNODE(os)->dn_zio = zio;
1330 dnode_sync(DMU_META_DNODE(os), tx);
1332 os->os_phys->os_flags = os->os_flags;
1334 if (DMU_USERUSED_DNODE(os) &&
1335 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1336 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1337 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1338 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1339 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1342 txgoff = tx->tx_txg & TXG_MASK;
1344 if (dmu_objset_userused_enabled(os)) {
1346 * We must create the list here because it uses the
1347 * dn_dirty_link[] of this txg. But it may already
1348 * exist because we call dsl_dataset_sync() twice per txg.
1350 if (os->os_synced_dnodes == NULL) {
1351 os->os_synced_dnodes =
1352 multilist_create(sizeof (dnode_t),
1353 offsetof(dnode_t, dn_dirty_link[txgoff]),
1354 dnode_multilist_index_func);
1356 ASSERT3U(os->os_synced_dnodes->ml_offset, ==,
1357 offsetof(dnode_t, dn_dirty_link[txgoff]));
1362 i < multilist_get_num_sublists(os->os_dirty_dnodes[txgoff]); i++) {
1363 sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP);
1364 sda->sda_list = os->os_dirty_dnodes[txgoff];
1365 sda->sda_sublist_idx = i;
1367 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
1368 sync_dnodes_task, sda, 0);
1369 /* callback frees sda */
1371 taskq_wait(dmu_objset_pool(os)->dp_sync_taskq);
1373 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1374 while ((dr = list_head(list)) != NULL) {
1375 ASSERT0(dr->dr_dbuf->db_level);
1376 list_remove(list, dr);
1378 zio_nowait(dr->dr_zio);
1381 /* Enable dnode backfill if enough objects have been freed. */
1382 if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) {
1383 os->os_rescan_dnodes = B_TRUE;
1384 os->os_freed_dnodes = 0;
1388 * Free intent log blocks up to this tx.
1390 zil_sync(os->os_zil, tx);
1391 os->os_phys->os_zil_header = os->os_zil_header;
1396 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1398 return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK]));
1401 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1404 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1410 dmu_objset_userused_enabled(objset_t *os)
1412 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1413 used_cbs[os->os_phys->os_type] != NULL &&
1414 DMU_USERUSED_DNODE(os) != NULL);
1417 typedef struct userquota_node {
1420 avl_node_t uqn_node;
1423 typedef struct userquota_cache {
1424 avl_tree_t uqc_user_deltas;
1425 avl_tree_t uqc_group_deltas;
1426 } userquota_cache_t;
1429 userquota_compare(const void *l, const void *r)
1431 const userquota_node_t *luqn = l;
1432 const userquota_node_t *ruqn = r;
1434 if (luqn->uqn_id < ruqn->uqn_id)
1436 if (luqn->uqn_id > ruqn->uqn_id)
1442 do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx)
1445 userquota_node_t *uqn;
1447 ASSERT(dmu_tx_is_syncing(tx));
1450 while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas,
1451 &cookie)) != NULL) {
1453 * os_userused_lock protects against concurrent calls to
1454 * zap_increment_int(). It's needed because zap_increment_int()
1455 * is not thread-safe (i.e. not atomic).
1457 mutex_enter(&os->os_userused_lock);
1458 VERIFY0(zap_increment_int(os, DMU_USERUSED_OBJECT,
1459 uqn->uqn_id, uqn->uqn_delta, tx));
1460 mutex_exit(&os->os_userused_lock);
1461 kmem_free(uqn, sizeof (*uqn));
1463 avl_destroy(&cache->uqc_user_deltas);
1466 while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas,
1467 &cookie)) != NULL) {
1468 mutex_enter(&os->os_userused_lock);
1469 VERIFY0(zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1470 uqn->uqn_id, uqn->uqn_delta, tx));
1471 mutex_exit(&os->os_userused_lock);
1472 kmem_free(uqn, sizeof (*uqn));
1474 avl_destroy(&cache->uqc_group_deltas);
1478 userquota_update_cache(avl_tree_t *avl, uint64_t id, int64_t delta)
1480 userquota_node_t search = { .uqn_id = id };
1483 userquota_node_t *uqn = avl_find(avl, &search, &idx);
1485 uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP);
1487 avl_insert(avl, uqn, idx);
1489 uqn->uqn_delta += delta;
1493 do_userquota_update(userquota_cache_t *cache, uint64_t used, uint64_t flags,
1494 uint64_t user, uint64_t group, boolean_t subtract)
1496 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1497 int64_t delta = DNODE_SIZE + used;
1501 userquota_update_cache(&cache->uqc_user_deltas, user, delta);
1502 userquota_update_cache(&cache->uqc_group_deltas, group, delta);
1506 typedef struct userquota_updates_arg {
1508 int uua_sublist_idx;
1510 } userquota_updates_arg_t;
1513 userquota_updates_task(void *arg)
1515 userquota_updates_arg_t *uua = arg;
1516 objset_t *os = uua->uua_os;
1517 dmu_tx_t *tx = uua->uua_tx;
1519 userquota_cache_t cache = { 0 };
1521 multilist_sublist_t *list =
1522 multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx);
1524 ASSERT(multilist_sublist_head(list) == NULL ||
1525 dmu_objset_userused_enabled(os));
1526 avl_create(&cache.uqc_user_deltas, userquota_compare,
1527 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1528 avl_create(&cache.uqc_group_deltas, userquota_compare,
1529 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1531 while ((dn = multilist_sublist_head(list)) != NULL) {
1533 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1534 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1535 dn->dn_phys->dn_flags &
1536 DNODE_FLAG_USERUSED_ACCOUNTED);
1538 flags = dn->dn_id_flags;
1540 if (flags & DN_ID_OLD_EXIST) {
1541 do_userquota_update(&cache,
1542 dn->dn_oldused, dn->dn_oldflags,
1543 dn->dn_olduid, dn->dn_oldgid, B_TRUE);
1545 if (flags & DN_ID_NEW_EXIST) {
1546 do_userquota_update(&cache,
1547 DN_USED_BYTES(dn->dn_phys),
1548 dn->dn_phys->dn_flags, dn->dn_newuid,
1549 dn->dn_newgid, B_FALSE);
1552 mutex_enter(&dn->dn_mtx);
1554 dn->dn_oldflags = 0;
1555 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1556 dn->dn_olduid = dn->dn_newuid;
1557 dn->dn_oldgid = dn->dn_newgid;
1558 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1559 if (dn->dn_bonuslen == 0)
1560 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1562 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1564 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1565 mutex_exit(&dn->dn_mtx);
1567 multilist_sublist_remove(list, dn);
1568 dnode_rele(dn, os->os_synced_dnodes);
1570 do_userquota_cacheflush(os, &cache, tx);
1571 multilist_sublist_unlock(list);
1572 kmem_free(uua, sizeof (*uua));
1576 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1578 if (!dmu_objset_userused_enabled(os))
1581 /* Allocate the user/groupused objects if necessary. */
1582 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1583 VERIFY0(zap_create_claim(os,
1584 DMU_USERUSED_OBJECT,
1585 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1586 VERIFY0(zap_create_claim(os,
1587 DMU_GROUPUSED_OBJECT,
1588 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1592 i < multilist_get_num_sublists(os->os_synced_dnodes); i++) {
1593 userquota_updates_arg_t *uua =
1594 kmem_alloc(sizeof (*uua), KM_SLEEP);
1596 uua->uua_sublist_idx = i;
1598 /* note: caller does taskq_wait() */
1599 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
1600 userquota_updates_task, uua, 0);
1601 /* callback frees uua */
1606 * Returns a pointer to data to find uid/gid from
1608 * If a dirty record for transaction group that is syncing can't
1609 * be found then NULL is returned. In the NULL case it is assumed
1610 * the uid/gid aren't changing.
1613 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1615 dbuf_dirty_record_t *dr, **drp;
1618 if (db->db_dirtycnt == 0)
1619 return (db->db.db_data); /* Nothing is changing */
1621 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1622 if (dr->dr_txg == tx->tx_txg)
1630 DB_DNODE_ENTER(dr->dr_dbuf);
1631 dn = DB_DNODE(dr->dr_dbuf);
1633 if (dn->dn_bonuslen == 0 &&
1634 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1635 data = dr->dt.dl.dr_data->b_data;
1637 data = dr->dt.dl.dr_data;
1639 DB_DNODE_EXIT(dr->dr_dbuf);
1646 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1648 objset_t *os = dn->dn_objset;
1650 dmu_buf_impl_t *db = NULL;
1651 uint64_t *user = NULL;
1652 uint64_t *group = NULL;
1653 int flags = dn->dn_id_flags;
1655 boolean_t have_spill = B_FALSE;
1657 if (!dmu_objset_userused_enabled(dn->dn_objset))
1660 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1661 DN_ID_CHKED_SPILL)))
1664 if (before && dn->dn_bonuslen != 0)
1665 data = DN_BONUS(dn->dn_phys);
1666 else if (!before && dn->dn_bonuslen != 0) {
1669 mutex_enter(&db->db_mtx);
1670 data = dmu_objset_userquota_find_data(db, tx);
1672 data = DN_BONUS(dn->dn_phys);
1674 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1677 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1678 rf |= DB_RF_HAVESTRUCT;
1679 error = dmu_spill_hold_by_dnode(dn,
1680 rf | DB_RF_MUST_SUCCEED,
1681 FTAG, (dmu_buf_t **)&db);
1683 mutex_enter(&db->db_mtx);
1684 data = (before) ? db->db.db_data :
1685 dmu_objset_userquota_find_data(db, tx);
1686 have_spill = B_TRUE;
1688 mutex_enter(&dn->dn_mtx);
1689 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1690 mutex_exit(&dn->dn_mtx);
1696 user = &dn->dn_olduid;
1697 group = &dn->dn_oldgid;
1699 user = &dn->dn_newuid;
1700 group = &dn->dn_newgid;
1704 * Must always call the callback in case the object
1705 * type has changed and that type isn't an object type to track
1707 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1711 * Preserve existing uid/gid when the callback can't determine
1712 * what the new uid/gid are and the callback returned EEXIST.
1713 * The EEXIST error tells us to just use the existing uid/gid.
1714 * If we don't know what the old values are then just assign
1715 * them to 0, since that is a new file being created.
1717 if (!before && data == NULL && error == EEXIST) {
1718 if (flags & DN_ID_OLD_EXIST) {
1719 dn->dn_newuid = dn->dn_olduid;
1720 dn->dn_newgid = dn->dn_oldgid;
1729 mutex_exit(&db->db_mtx);
1731 mutex_enter(&dn->dn_mtx);
1732 if (error == 0 && before)
1733 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1734 if (error == 0 && !before)
1735 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1738 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1740 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1742 mutex_exit(&dn->dn_mtx);
1744 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1748 dmu_objset_userspace_present(objset_t *os)
1750 return (os->os_phys->os_flags &
1751 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1755 dmu_objset_userspace_upgrade(objset_t *os)
1760 if (dmu_objset_userspace_present(os))
1762 if (!dmu_objset_userused_enabled(os))
1763 return (SET_ERROR(ENOTSUP));
1764 if (dmu_objset_is_snapshot(os))
1765 return (SET_ERROR(EINVAL));
1768 * We simply need to mark every object dirty, so that it will be
1769 * synced out and now accounted. If this is called
1770 * concurrently, or if we already did some work before crashing,
1771 * that's fine, since we track each object's accounted state
1775 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1780 if (issig(JUSTLOOKING) && issig(FORREAL))
1781 return (SET_ERROR(EINTR));
1783 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1786 tx = dmu_tx_create(os);
1787 dmu_tx_hold_bonus(tx, obj);
1788 objerr = dmu_tx_assign(tx, TXG_WAIT);
1793 dmu_buf_will_dirty(db, tx);
1794 dmu_buf_rele(db, FTAG);
1798 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1799 txg_wait_synced(dmu_objset_pool(os), 0);
1804 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1805 uint64_t *usedobjsp, uint64_t *availobjsp)
1807 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1808 usedobjsp, availobjsp);
1812 dmu_objset_fsid_guid(objset_t *os)
1814 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1818 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1820 stat->dds_type = os->os_phys->os_type;
1821 if (os->os_dsl_dataset)
1822 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1826 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1828 ASSERT(os->os_dsl_dataset ||
1829 os->os_phys->os_type == DMU_OST_META);
1831 if (os->os_dsl_dataset != NULL)
1832 dsl_dataset_stats(os->os_dsl_dataset, nv);
1834 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1835 os->os_phys->os_type);
1836 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1837 dmu_objset_userspace_present(os));
1841 dmu_objset_is_snapshot(objset_t *os)
1843 if (os->os_dsl_dataset != NULL)
1844 return (os->os_dsl_dataset->ds_is_snapshot);
1850 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1851 boolean_t *conflict)
1853 dsl_dataset_t *ds = os->os_dsl_dataset;
1856 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1857 return (SET_ERROR(ENOENT));
1859 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1860 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
1861 MT_NORMALIZE, real, maxlen, conflict));
1865 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1866 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1868 dsl_dataset_t *ds = os->os_dsl_dataset;
1869 zap_cursor_t cursor;
1870 zap_attribute_t attr;
1872 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1874 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1875 return (SET_ERROR(ENOENT));
1877 zap_cursor_init_serialized(&cursor,
1878 ds->ds_dir->dd_pool->dp_meta_objset,
1879 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
1881 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1882 zap_cursor_fini(&cursor);
1883 return (SET_ERROR(ENOENT));
1886 if (strlen(attr.za_name) + 1 > namelen) {
1887 zap_cursor_fini(&cursor);
1888 return (SET_ERROR(ENAMETOOLONG));
1891 (void) strcpy(name, attr.za_name);
1893 *idp = attr.za_first_integer;
1895 *case_conflict = attr.za_normalization_conflict;
1896 zap_cursor_advance(&cursor);
1897 *offp = zap_cursor_serialize(&cursor);
1898 zap_cursor_fini(&cursor);
1904 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1905 uint64_t *idp, uint64_t *offp)
1907 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1908 zap_cursor_t cursor;
1909 zap_attribute_t attr;
1911 /* there is no next dir on a snapshot! */
1912 if (os->os_dsl_dataset->ds_object !=
1913 dsl_dir_phys(dd)->dd_head_dataset_obj)
1914 return (SET_ERROR(ENOENT));
1916 zap_cursor_init_serialized(&cursor,
1917 dd->dd_pool->dp_meta_objset,
1918 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
1920 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1921 zap_cursor_fini(&cursor);
1922 return (SET_ERROR(ENOENT));
1925 if (strlen(attr.za_name) + 1 > namelen) {
1926 zap_cursor_fini(&cursor);
1927 return (SET_ERROR(ENAMETOOLONG));
1930 (void) strcpy(name, attr.za_name);
1932 *idp = attr.za_first_integer;
1933 zap_cursor_advance(&cursor);
1934 *offp = zap_cursor_serialize(&cursor);
1935 zap_cursor_fini(&cursor);
1940 typedef struct dmu_objset_find_ctx {
1944 char *dc_ddname; /* last component of ddobj's name */
1945 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *);
1948 kmutex_t *dc_error_lock;
1950 } dmu_objset_find_ctx_t;
1953 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp)
1955 dsl_pool_t *dp = dcp->dc_dp;
1959 zap_attribute_t *attr;
1963 /* don't process if there already was an error */
1964 if (*dcp->dc_error != 0)
1968 * Note: passing the name (dc_ddname) here is optional, but it
1969 * improves performance because we don't need to call
1970 * zap_value_search() to determine the name.
1972 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd);
1976 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1977 if (dd->dd_myname[0] == '$') {
1978 dsl_dir_rele(dd, FTAG);
1982 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1983 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1986 * Iterate over all children.
1988 if (dcp->dc_flags & DS_FIND_CHILDREN) {
1989 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1990 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1991 zap_cursor_retrieve(&zc, attr) == 0;
1992 (void) zap_cursor_advance(&zc)) {
1993 ASSERT3U(attr->za_integer_length, ==,
1995 ASSERT3U(attr->za_num_integers, ==, 1);
1997 dmu_objset_find_ctx_t *child_dcp =
1998 kmem_alloc(sizeof (*child_dcp), KM_SLEEP);
2000 child_dcp->dc_ddobj = attr->za_first_integer;
2001 child_dcp->dc_ddname = spa_strdup(attr->za_name);
2002 if (dcp->dc_tq != NULL)
2003 (void) taskq_dispatch(dcp->dc_tq,
2004 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP);
2006 dmu_objset_find_dp_impl(child_dcp);
2008 zap_cursor_fini(&zc);
2012 * Iterate over all snapshots.
2014 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) {
2016 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2021 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2022 dsl_dataset_rele(ds, FTAG);
2024 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2025 zap_cursor_retrieve(&zc, attr) == 0;
2026 (void) zap_cursor_advance(&zc)) {
2027 ASSERT3U(attr->za_integer_length, ==,
2029 ASSERT3U(attr->za_num_integers, ==, 1);
2031 err = dsl_dataset_hold_obj(dp,
2032 attr->za_first_integer, FTAG, &ds);
2035 err = dcp->dc_func(dp, ds, dcp->dc_arg);
2036 dsl_dataset_rele(ds, FTAG);
2040 zap_cursor_fini(&zc);
2044 kmem_free(attr, sizeof (zap_attribute_t));
2047 dsl_dir_rele(dd, FTAG);
2054 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2057 * Note: we hold the dir while calling dsl_dataset_hold_obj() so
2058 * that the dir will remain cached, and we won't have to re-instantiate
2059 * it (which could be expensive due to finding its name via
2060 * zap_value_search()).
2062 dsl_dir_rele(dd, FTAG);
2065 err = dcp->dc_func(dp, ds, dcp->dc_arg);
2066 dsl_dataset_rele(ds, FTAG);
2070 mutex_enter(dcp->dc_error_lock);
2071 /* only keep first error */
2072 if (*dcp->dc_error == 0)
2073 *dcp->dc_error = err;
2074 mutex_exit(dcp->dc_error_lock);
2077 if (dcp->dc_ddname != NULL)
2078 spa_strfree(dcp->dc_ddname);
2079 kmem_free(dcp, sizeof (*dcp));
2083 dmu_objset_find_dp_cb(void *arg)
2085 dmu_objset_find_ctx_t *dcp = arg;
2086 dsl_pool_t *dp = dcp->dc_dp;
2089 * We need to get a pool_config_lock here, as there are several
2090 * asssert(pool_config_held) down the stack. Getting a lock via
2091 * dsl_pool_config_enter is risky, as it might be stalled by a
2092 * pending writer. This would deadlock, as the write lock can
2093 * only be granted when our parent thread gives up the lock.
2094 * The _prio interface gives us priority over a pending writer.
2096 dsl_pool_config_enter_prio(dp, FTAG);
2098 dmu_objset_find_dp_impl(dcp);
2100 dsl_pool_config_exit(dp, FTAG);
2104 * Find objsets under and including ddobj, call func(ds) on each.
2105 * The order for the enumeration is completely undefined.
2106 * func is called with dsl_pool_config held.
2109 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
2110 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
2115 dmu_objset_find_ctx_t *dcp;
2118 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL);
2119 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP);
2122 dcp->dc_ddobj = ddobj;
2123 dcp->dc_ddname = NULL;
2124 dcp->dc_func = func;
2126 dcp->dc_flags = flags;
2127 dcp->dc_error_lock = &err_lock;
2128 dcp->dc_error = &error;
2130 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) {
2132 * In case a write lock is held we can't make use of
2133 * parallelism, as down the stack of the worker threads
2134 * the lock is asserted via dsl_pool_config_held.
2135 * In case of a read lock this is solved by getting a read
2136 * lock in each worker thread, which isn't possible in case
2137 * of a writer lock. So we fall back to the synchronous path
2139 * In the future it might be possible to get some magic into
2140 * dsl_pool_config_held in a way that it returns true for
2141 * the worker threads so that a single lock held from this
2142 * thread suffices. For now, stay single threaded.
2144 dmu_objset_find_dp_impl(dcp);
2145 mutex_destroy(&err_lock);
2150 ntasks = dmu_find_threads;
2152 ntasks = vdev_count_leaves(dp->dp_spa) * 4;
2153 tq = taskq_create("dmu_objset_find", ntasks, minclsyspri, ntasks,
2156 kmem_free(dcp, sizeof (*dcp));
2157 mutex_destroy(&err_lock);
2159 return (SET_ERROR(ENOMEM));
2163 /* dcp will be freed by task */
2164 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP);
2167 * PORTING: this code relies on the property of taskq_wait to wait
2168 * until no more tasks are queued and no more tasks are active. As
2169 * we always queue new tasks from within other tasks, task_wait
2170 * reliably waits for the full recursion to finish, even though we
2171 * enqueue new tasks after taskq_wait has been called.
2172 * On platforms other than illumos, taskq_wait may not have this
2177 mutex_destroy(&err_lock);
2183 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
2184 * The dp_config_rwlock must not be held when this is called, and it
2185 * will not be held when the callback is called.
2186 * Therefore this function should only be used when the pool is not changing
2187 * (e.g. in syncing context), or the callback can deal with the possible races.
2190 dmu_objset_find_impl(spa_t *spa, const char *name,
2191 int func(const char *, void *), void *arg, int flags)
2194 dsl_pool_t *dp = spa_get_dsl(spa);
2197 zap_attribute_t *attr;
2202 dsl_pool_config_enter(dp, FTAG);
2204 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
2206 dsl_pool_config_exit(dp, FTAG);
2210 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2211 if (dd->dd_myname[0] == '$') {
2212 dsl_dir_rele(dd, FTAG);
2213 dsl_pool_config_exit(dp, FTAG);
2217 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
2218 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
2221 * Iterate over all children.
2223 if (flags & DS_FIND_CHILDREN) {
2224 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2225 dsl_dir_phys(dd)->dd_child_dir_zapobj);
2226 zap_cursor_retrieve(&zc, attr) == 0;
2227 (void) zap_cursor_advance(&zc)) {
2228 ASSERT3U(attr->za_integer_length, ==,
2230 ASSERT3U(attr->za_num_integers, ==, 1);
2232 child = kmem_asprintf("%s/%s", name, attr->za_name);
2233 dsl_pool_config_exit(dp, FTAG);
2234 err = dmu_objset_find_impl(spa, child,
2236 dsl_pool_config_enter(dp, FTAG);
2241 zap_cursor_fini(&zc);
2244 dsl_dir_rele(dd, FTAG);
2245 dsl_pool_config_exit(dp, FTAG);
2246 kmem_free(attr, sizeof (zap_attribute_t));
2252 * Iterate over all snapshots.
2254 if (flags & DS_FIND_SNAPSHOTS) {
2255 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2260 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2261 dsl_dataset_rele(ds, FTAG);
2263 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2264 zap_cursor_retrieve(&zc, attr) == 0;
2265 (void) zap_cursor_advance(&zc)) {
2266 ASSERT3U(attr->za_integer_length, ==,
2268 ASSERT3U(attr->za_num_integers, ==, 1);
2270 child = kmem_asprintf("%s@%s",
2271 name, attr->za_name);
2272 dsl_pool_config_exit(dp, FTAG);
2273 err = func(child, arg);
2274 dsl_pool_config_enter(dp, FTAG);
2279 zap_cursor_fini(&zc);
2283 dsl_dir_rele(dd, FTAG);
2284 kmem_free(attr, sizeof (zap_attribute_t));
2285 dsl_pool_config_exit(dp, FTAG);
2290 /* Apply to self. */
2291 return (func(name, arg));
2295 * See comment above dmu_objset_find_impl().
2298 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
2304 error = spa_open(name, &spa, FTAG);
2307 error = dmu_objset_find_impl(spa, name, func, arg, flags);
2308 spa_close(spa, FTAG);
2313 dmu_objset_set_user(objset_t *os, void *user_ptr)
2315 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2316 os->os_user_ptr = user_ptr;
2320 dmu_objset_get_user(objset_t *os)
2322 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2323 return (os->os_user_ptr);
2327 * Determine name of filesystem, given name of snapshot.
2328 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
2331 dmu_fsname(const char *snapname, char *buf)
2333 char *atp = strchr(snapname, '@');
2335 return (SET_ERROR(EINVAL));
2336 if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN)
2337 return (SET_ERROR(ENAMETOOLONG));
2338 (void) strlcpy(buf, snapname, atp - snapname + 1);
2343 * Call when we think we're going to write/free space in open context to track
2344 * the amount of dirty data in the open txg, which is also the amount
2345 * of memory that can not be evicted until this txg syncs.
2348 dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx)
2350 dsl_dataset_t *ds = os->os_dsl_dataset;
2351 int64_t aspace = spa_get_worst_case_asize(os->os_spa, space);
2354 dsl_dir_willuse_space(ds->ds_dir, aspace, tx);
2355 dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx);