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>
57 #include "zfs_namecheck.h"
60 * Needed to close a window in dnode_move() that allows the objset to be freed
61 * before it can be safely accessed.
66 * Tunable to overwrite the maximum number of threads for the parallization
67 * of dmu_objset_find_dp, needed to speed up the import of pools with many
69 * Default is 4 times the number of leaf vdevs.
71 int dmu_find_threads = 0;
74 * Backfill lower metadnode objects after this many have been freed.
75 * Backfilling negatively impacts object creation rates, so only do it
76 * if there are enough holes to fill.
78 int dmu_rescan_dnode_threshold = 131072;
80 static void dmu_objset_find_dp_cb(void *arg);
85 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
95 dmu_objset_spa(objset_t *os)
101 dmu_objset_zil(objset_t *os)
107 dmu_objset_pool(objset_t *os)
111 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
112 return (ds->ds_dir->dd_pool);
114 return (spa_get_dsl(os->os_spa));
118 dmu_objset_ds(objset_t *os)
120 return (os->os_dsl_dataset);
124 dmu_objset_type(objset_t *os)
126 return (os->os_phys->os_type);
130 dmu_objset_name(objset_t *os, char *buf)
132 dsl_dataset_name(os->os_dsl_dataset, buf);
136 dmu_objset_id(objset_t *os)
138 dsl_dataset_t *ds = os->os_dsl_dataset;
140 return (ds ? ds->ds_object : 0);
144 dmu_objset_dnodesize(objset_t *os)
146 return (os->os_dnodesize);
150 dmu_objset_syncprop(objset_t *os)
152 return (os->os_sync);
156 dmu_objset_logbias(objset_t *os)
158 return (os->os_logbias);
162 checksum_changed_cb(void *arg, uint64_t newval)
167 * Inheritance should have been done by now.
169 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
171 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
175 compression_changed_cb(void *arg, uint64_t newval)
180 * Inheritance and range checking should have been done by now.
182 ASSERT(newval != ZIO_COMPRESS_INHERIT);
184 os->os_compress = zio_compress_select(os->os_spa, newval,
189 copies_changed_cb(void *arg, uint64_t newval)
194 * Inheritance and range checking should have been done by now.
197 ASSERT(newval <= spa_max_replication(os->os_spa));
199 os->os_copies = newval;
203 dedup_changed_cb(void *arg, uint64_t newval)
206 spa_t *spa = os->os_spa;
207 enum zio_checksum checksum;
210 * Inheritance should have been done by now.
212 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
214 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
216 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
217 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
221 primary_cache_changed_cb(void *arg, uint64_t newval)
226 * Inheritance and range checking should have been done by now.
228 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
229 newval == ZFS_CACHE_METADATA);
231 os->os_primary_cache = newval;
235 secondary_cache_changed_cb(void *arg, uint64_t newval)
240 * Inheritance and range checking should have been done by now.
242 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
243 newval == ZFS_CACHE_METADATA);
245 os->os_secondary_cache = newval;
249 sync_changed_cb(void *arg, uint64_t newval)
254 * Inheritance and range checking should have been done by now.
256 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
257 newval == ZFS_SYNC_DISABLED);
259 os->os_sync = newval;
261 zil_set_sync(os->os_zil, newval);
265 redundant_metadata_changed_cb(void *arg, uint64_t newval)
270 * Inheritance and range checking should have been done by now.
272 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
273 newval == ZFS_REDUNDANT_METADATA_MOST);
275 os->os_redundant_metadata = newval;
279 dnodesize_changed_cb(void *arg, uint64_t newval)
284 case ZFS_DNSIZE_LEGACY:
285 os->os_dnodesize = DNODE_MIN_SIZE;
287 case ZFS_DNSIZE_AUTO:
289 * Choose a dnode size that will work well for most
290 * workloads if the user specified "auto". Future code
291 * improvements could dynamically select a dnode size
292 * based on observed workload patterns.
294 os->os_dnodesize = DNODE_MIN_SIZE * 2;
301 os->os_dnodesize = newval;
307 logbias_changed_cb(void *arg, uint64_t newval)
311 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
312 newval == ZFS_LOGBIAS_THROUGHPUT);
313 os->os_logbias = newval;
315 zil_set_logbias(os->os_zil, newval);
319 recordsize_changed_cb(void *arg, uint64_t newval)
323 os->os_recordsize = newval;
327 dmu_objset_byteswap(void *buf, size_t size)
329 objset_phys_t *osp = buf;
331 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
332 dnode_byteswap(&osp->os_meta_dnode);
333 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
334 osp->os_type = BSWAP_64(osp->os_type);
335 osp->os_flags = BSWAP_64(osp->os_flags);
336 if (size == sizeof (objset_phys_t)) {
337 dnode_byteswap(&osp->os_userused_dnode);
338 dnode_byteswap(&osp->os_groupused_dnode);
343 * The hash is a CRC-based hash of the objset_t pointer and the object number.
346 dnode_hash(const objset_t *os, uint64_t obj)
348 uintptr_t osv = (uintptr_t)os;
349 uint64_t crc = -1ULL;
351 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
353 * The low 6 bits of the pointer don't have much entropy, because
354 * the objset_t is larger than 2^6 bytes long.
356 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF];
357 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF];
358 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF];
359 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF];
361 crc ^= (osv>>14) ^ (obj>>24);
367 dnode_multilist_index_func(multilist_t *ml, void *obj)
370 return (dnode_hash(dn->dn_objset, dn->dn_object) %
371 multilist_get_num_sublists(ml));
375 * Instantiates the objset_t in-memory structure corresponding to the
376 * objset_phys_t that's pointed to by the specified blkptr_t.
379 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
385 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
389 * The $ORIGIN dataset (if it exists) doesn't have an associated
390 * objset, so there's no reason to open it. The $ORIGIN dataset
391 * will not exist on pools older than SPA_VERSION_ORIGIN.
393 if (ds != NULL && spa_get_dsl(spa) != NULL &&
394 spa_get_dsl(spa)->dp_origin_snap != NULL) {
395 ASSERT3P(ds->ds_dir, !=,
396 spa_get_dsl(spa)->dp_origin_snap->ds_dir);
400 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
401 os->os_dsl_dataset = ds;
404 if (!BP_IS_HOLE(os->os_rootbp)) {
405 arc_flags_t aflags = ARC_FLAG_WAIT;
407 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
408 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
410 if (DMU_OS_IS_L2CACHEABLE(os))
411 aflags |= ARC_FLAG_L2CACHE;
413 dprintf_bp(os->os_rootbp, "reading %s", "");
414 err = arc_read(NULL, spa, os->os_rootbp,
415 arc_getbuf_func, &os->os_phys_buf,
416 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
418 kmem_free(os, sizeof (objset_t));
419 /* convert checksum errors into IO errors */
421 err = SET_ERROR(EIO);
425 /* Increase the blocksize if we are permitted. */
426 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
427 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
428 arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf,
429 ARC_BUFC_METADATA, sizeof (objset_phys_t));
430 bzero(buf->b_data, sizeof (objset_phys_t));
431 bcopy(os->os_phys_buf->b_data, buf->b_data,
432 arc_buf_size(os->os_phys_buf));
433 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
434 os->os_phys_buf = buf;
437 os->os_phys = os->os_phys_buf->b_data;
438 os->os_flags = os->os_phys->os_flags;
440 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
441 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
442 os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf,
443 ARC_BUFC_METADATA, size);
444 os->os_phys = os->os_phys_buf->b_data;
445 bzero(os->os_phys, size);
449 * Note: the changed_cb will be called once before the register
450 * func returns, thus changing the checksum/compression from the
451 * default (fletcher2/off). Snapshots don't need to know about
452 * checksum/compression/copies.
455 boolean_t needlock = B_FALSE;
458 * Note: it's valid to open the objset if the dataset is
459 * long-held, in which case the pool_config lock will not
462 if (!dsl_pool_config_held(dmu_objset_pool(os))) {
464 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
466 err = dsl_prop_register(ds,
467 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
468 primary_cache_changed_cb, os);
470 err = dsl_prop_register(ds,
471 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
472 secondary_cache_changed_cb, os);
474 if (!ds->ds_is_snapshot) {
476 err = dsl_prop_register(ds,
477 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
478 checksum_changed_cb, os);
481 err = dsl_prop_register(ds,
482 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
483 compression_changed_cb, os);
486 err = dsl_prop_register(ds,
487 zfs_prop_to_name(ZFS_PROP_COPIES),
488 copies_changed_cb, os);
491 err = dsl_prop_register(ds,
492 zfs_prop_to_name(ZFS_PROP_DEDUP),
493 dedup_changed_cb, os);
496 err = dsl_prop_register(ds,
497 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
498 logbias_changed_cb, os);
501 err = dsl_prop_register(ds,
502 zfs_prop_to_name(ZFS_PROP_SYNC),
503 sync_changed_cb, os);
506 err = dsl_prop_register(ds,
508 ZFS_PROP_REDUNDANT_METADATA),
509 redundant_metadata_changed_cb, os);
512 err = dsl_prop_register(ds,
513 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
514 recordsize_changed_cb, os);
517 err = dsl_prop_register(ds,
518 zfs_prop_to_name(ZFS_PROP_DNODESIZE),
519 dnodesize_changed_cb, os);
523 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
525 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
526 kmem_free(os, sizeof (objset_t));
530 /* It's the meta-objset. */
531 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
532 os->os_compress = ZIO_COMPRESS_ON;
533 os->os_copies = spa_max_replication(spa);
534 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
535 os->os_dedup_verify = B_FALSE;
536 os->os_logbias = ZFS_LOGBIAS_LATENCY;
537 os->os_sync = ZFS_SYNC_STANDARD;
538 os->os_primary_cache = ZFS_CACHE_ALL;
539 os->os_secondary_cache = ZFS_CACHE_ALL;
540 os->os_dnodesize = DNODE_MIN_SIZE;
543 * These properties will be filled in by the logic in zfs_get_zplprop()
544 * when they are queried for the first time.
546 os->os_version = OBJSET_PROP_UNINITIALIZED;
547 os->os_normalization = OBJSET_PROP_UNINITIALIZED;
548 os->os_utf8only = OBJSET_PROP_UNINITIALIZED;
549 os->os_casesensitivity = OBJSET_PROP_UNINITIALIZED;
551 if (ds == NULL || !ds->ds_is_snapshot)
552 os->os_zil_header = os->os_phys->os_zil_header;
553 os->os_zil = zil_alloc(os, &os->os_zil_header);
555 for (i = 0; i < TXG_SIZE; i++) {
556 os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t),
557 offsetof(dnode_t, dn_dirty_link[i]),
558 dnode_multilist_index_func);
560 list_create(&os->os_dnodes, sizeof (dnode_t),
561 offsetof(dnode_t, dn_link));
562 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
563 offsetof(dmu_buf_impl_t, db_link));
565 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
566 mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL);
567 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
568 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
570 dnode_special_open(os, &os->os_phys->os_meta_dnode,
571 DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
572 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
573 dnode_special_open(os, &os->os_phys->os_userused_dnode,
574 DMU_USERUSED_OBJECT, &os->os_userused_dnode);
575 dnode_special_open(os, &os->os_phys->os_groupused_dnode,
576 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
584 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
589 * We shouldn't be doing anything with dsl_dataset_t's unless the
590 * pool_config lock is held, or the dataset is long-held.
592 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) ||
593 dsl_dataset_long_held(ds));
595 mutex_enter(&ds->ds_opening_lock);
596 if (ds->ds_objset == NULL) {
598 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
599 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
600 ds, dsl_dataset_get_blkptr(ds), &os);
601 rrw_exit(&ds->ds_bp_rwlock, FTAG);
604 mutex_enter(&ds->ds_lock);
605 ASSERT(ds->ds_objset == NULL);
607 mutex_exit(&ds->ds_lock);
610 *osp = ds->ds_objset;
611 mutex_exit(&ds->ds_opening_lock);
616 * Holds the pool while the objset is held. Therefore only one objset
617 * can be held at a time.
620 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
626 err = dsl_pool_hold(name, tag, &dp);
629 err = dsl_dataset_hold(dp, name, tag, &ds);
631 dsl_pool_rele(dp, tag);
635 err = dmu_objset_from_ds(ds, osp);
637 dsl_dataset_rele(ds, tag);
638 dsl_pool_rele(dp, tag);
645 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type,
646 boolean_t readonly, void *tag, objset_t **osp)
650 err = dmu_objset_from_ds(ds, osp);
652 dsl_dataset_disown(ds, tag);
653 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
654 dsl_dataset_disown(ds, tag);
655 return (SET_ERROR(EINVAL));
656 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
657 dsl_dataset_disown(ds, tag);
658 return (SET_ERROR(EROFS));
664 * dsl_pool must not be held when this is called.
665 * Upon successful return, there will be a longhold on the dataset,
666 * and the dsl_pool will not be held.
669 dmu_objset_own(const char *name, dmu_objset_type_t type,
670 boolean_t readonly, void *tag, objset_t **osp)
676 err = dsl_pool_hold(name, FTAG, &dp);
679 err = dsl_dataset_own(dp, name, tag, &ds);
681 dsl_pool_rele(dp, FTAG);
684 err = dmu_objset_own_impl(ds, type, readonly, tag, osp);
685 dsl_pool_rele(dp, FTAG);
691 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type,
692 boolean_t readonly, void *tag, objset_t **osp)
697 err = dsl_dataset_own_obj(dp, obj, tag, &ds);
701 return (dmu_objset_own_impl(ds, type, readonly, tag, osp));
705 dmu_objset_rele(objset_t *os, void *tag)
707 dsl_pool_t *dp = dmu_objset_pool(os);
708 dsl_dataset_rele(os->os_dsl_dataset, tag);
709 dsl_pool_rele(dp, tag);
713 * When we are called, os MUST refer to an objset associated with a dataset
714 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
715 * == tag. We will then release and reacquire ownership of the dataset while
716 * holding the pool config_rwlock to avoid intervening namespace or ownership
719 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
720 * release the hold on its dataset and acquire a new one on the dataset of the
721 * same name so that it can be partially torn down and reconstructed.
724 dmu_objset_refresh_ownership(dsl_dataset_t *ds, dsl_dataset_t **newds,
728 char name[ZFS_MAX_DATASET_NAME_LEN];
730 VERIFY3P(ds, !=, NULL);
731 VERIFY3P(ds->ds_owner, ==, tag);
732 VERIFY(dsl_dataset_long_held(ds));
734 dsl_dataset_name(ds, name);
735 dp = ds->ds_dir->dd_pool;
736 dsl_pool_config_enter(dp, FTAG);
737 dsl_dataset_disown(ds, tag);
738 VERIFY0(dsl_dataset_own(dp, name, tag, newds));
739 dsl_pool_config_exit(dp, FTAG);
743 dmu_objset_disown(objset_t *os, void *tag)
745 dsl_dataset_disown(os->os_dsl_dataset, tag);
749 dmu_objset_evict_dbufs(objset_t *os)
754 mutex_enter(&os->os_lock);
755 dn = list_head(&os->os_dnodes);
758 * Skip dnodes without holds. We have to do this dance
759 * because dnode_add_ref() only works if there is already a
760 * hold. If the dnode has no holds, then it has no dbufs.
762 if (dnode_add_ref(dn, FTAG)) {
763 list_insert_after(&os->os_dnodes, dn, &dn_marker);
764 mutex_exit(&os->os_lock);
766 dnode_evict_dbufs(dn);
767 dnode_rele(dn, FTAG);
769 mutex_enter(&os->os_lock);
770 dn = list_next(&os->os_dnodes, &dn_marker);
771 list_remove(&os->os_dnodes, &dn_marker);
773 dn = list_next(&os->os_dnodes, dn);
776 mutex_exit(&os->os_lock);
778 if (DMU_USERUSED_DNODE(os) != NULL) {
779 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
780 dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
782 dnode_evict_dbufs(DMU_META_DNODE(os));
786 * Objset eviction processing is split into into two pieces.
787 * The first marks the objset as evicting, evicts any dbufs that
788 * have a refcount of zero, and then queues up the objset for the
789 * second phase of eviction. Once os->os_dnodes has been cleared by
790 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
791 * The second phase closes the special dnodes, dequeues the objset from
792 * the list of those undergoing eviction, and finally frees the objset.
794 * NOTE: Due to asynchronous eviction processing (invocation of
795 * dnode_buf_pageout()), it is possible for the meta dnode for the
796 * objset to have no holds even though os->os_dnodes is not empty.
799 dmu_objset_evict(objset_t *os)
801 dsl_dataset_t *ds = os->os_dsl_dataset;
803 for (int t = 0; t < TXG_SIZE; t++)
804 ASSERT(!dmu_objset_is_dirty(os, t));
807 dsl_prop_unregister_all(ds, os);
812 dmu_objset_evict_dbufs(os);
814 mutex_enter(&os->os_lock);
815 spa_evicting_os_register(os->os_spa, os);
816 if (list_is_empty(&os->os_dnodes)) {
817 mutex_exit(&os->os_lock);
818 dmu_objset_evict_done(os);
820 mutex_exit(&os->os_lock);
825 dmu_objset_evict_done(objset_t *os)
827 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
829 dnode_special_close(&os->os_meta_dnode);
830 if (DMU_USERUSED_DNODE(os)) {
831 dnode_special_close(&os->os_userused_dnode);
832 dnode_special_close(&os->os_groupused_dnode);
834 zil_free(os->os_zil);
836 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
839 * This is a barrier to prevent the objset from going away in
840 * dnode_move() until we can safely ensure that the objset is still in
841 * use. We consider the objset valid before the barrier and invalid
844 rw_enter(&os_lock, RW_READER);
847 mutex_destroy(&os->os_lock);
848 mutex_destroy(&os->os_userused_lock);
849 mutex_destroy(&os->os_obj_lock);
850 mutex_destroy(&os->os_user_ptr_lock);
851 for (int i = 0; i < TXG_SIZE; i++) {
852 multilist_destroy(os->os_dirty_dnodes[i]);
854 spa_evicting_os_deregister(os->os_spa, os);
855 kmem_free(os, sizeof (objset_t));
859 dmu_objset_snap_cmtime(objset_t *os)
861 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
864 /* called from dsl for meta-objset */
866 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
867 dmu_objset_type_t type, dmu_tx_t *tx)
872 ASSERT(dmu_tx_is_syncing(tx));
875 VERIFY0(dmu_objset_from_ds(ds, &os));
877 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
879 mdn = DMU_META_DNODE(os);
881 dnode_allocate(mdn, DMU_OT_DNODE, DNODE_BLOCK_SIZE, DN_MAX_INDBLKSHIFT,
882 DMU_OT_NONE, 0, DNODE_MIN_SLOTS, tx);
885 * We don't want to have to increase the meta-dnode's nlevels
886 * later, because then we could do it in quescing context while
887 * we are also accessing it in open context.
889 * This precaution is not necessary for the MOS (ds == NULL),
890 * because the MOS is only updated in syncing context.
891 * This is most fortunate: the MOS is the only objset that
892 * needs to be synced multiple times as spa_sync() iterates
893 * to convergence, so minimizing its dn_nlevels matters.
899 * Determine the number of levels necessary for the meta-dnode
900 * to contain DN_MAX_OBJECT dnodes. Note that in order to
901 * ensure that we do not overflow 64 bits, there has to be
902 * a nlevels that gives us a number of blocks > DN_MAX_OBJECT
903 * but < 2^64. Therefore,
904 * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) (10) must be
905 * less than (64 - log2(DN_MAX_OBJECT)) (16).
907 while ((uint64_t)mdn->dn_nblkptr <<
908 (mdn->dn_datablkshift - DNODE_SHIFT +
909 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
913 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
914 mdn->dn_nlevels = levels;
917 ASSERT(type != DMU_OST_NONE);
918 ASSERT(type != DMU_OST_ANY);
919 ASSERT(type < DMU_OST_NUMTYPES);
920 os->os_phys->os_type = type;
921 if (dmu_objset_userused_enabled(os)) {
922 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
923 os->os_flags = os->os_phys->os_flags;
926 dsl_dataset_dirty(ds, tx);
931 typedef struct dmu_objset_create_arg {
932 const char *doca_name;
934 void (*doca_userfunc)(objset_t *os, void *arg,
935 cred_t *cr, dmu_tx_t *tx);
937 dmu_objset_type_t doca_type;
939 } dmu_objset_create_arg_t;
943 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
945 dmu_objset_create_arg_t *doca = arg;
946 dsl_pool_t *dp = dmu_tx_pool(tx);
951 if (strchr(doca->doca_name, '@') != NULL)
952 return (SET_ERROR(EINVAL));
954 if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN)
955 return (SET_ERROR(ENAMETOOLONG));
957 if (dataset_nestcheck(doca->doca_name) != 0)
958 return (SET_ERROR(ENAMETOOLONG));
960 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
964 dsl_dir_rele(pdd, FTAG);
965 return (SET_ERROR(EEXIST));
967 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
969 dsl_dir_rele(pdd, FTAG);
975 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
977 dmu_objset_create_arg_t *doca = arg;
978 dsl_pool_t *dp = dmu_tx_pool(tx);
986 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
988 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
989 doca->doca_cred, tx);
991 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
992 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
993 bp = dsl_dataset_get_blkptr(ds);
994 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
995 ds, bp, doca->doca_type, tx);
996 rrw_exit(&ds->ds_bp_rwlock, FTAG);
998 if (doca->doca_userfunc != NULL) {
999 doca->doca_userfunc(os, doca->doca_userarg,
1000 doca->doca_cred, tx);
1003 spa_history_log_internal_ds(ds, "create", tx, "");
1004 dsl_dataset_rele(ds, FTAG);
1005 dsl_dir_rele(pdd, FTAG);
1009 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
1010 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
1012 dmu_objset_create_arg_t doca;
1014 doca.doca_name = name;
1015 doca.doca_cred = CRED();
1016 doca.doca_flags = flags;
1017 doca.doca_userfunc = func;
1018 doca.doca_userarg = arg;
1019 doca.doca_type = type;
1021 return (dsl_sync_task(name,
1022 dmu_objset_create_check, dmu_objset_create_sync, &doca,
1023 5, ZFS_SPACE_CHECK_NORMAL));
1026 typedef struct dmu_objset_clone_arg {
1027 const char *doca_clone;
1028 const char *doca_origin;
1030 } dmu_objset_clone_arg_t;
1034 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
1036 dmu_objset_clone_arg_t *doca = arg;
1040 dsl_dataset_t *origin;
1041 dsl_pool_t *dp = dmu_tx_pool(tx);
1043 if (strchr(doca->doca_clone, '@') != NULL)
1044 return (SET_ERROR(EINVAL));
1046 if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN)
1047 return (SET_ERROR(ENAMETOOLONG));
1049 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
1053 dsl_dir_rele(pdd, FTAG);
1054 return (SET_ERROR(EEXIST));
1057 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
1060 dsl_dir_rele(pdd, FTAG);
1061 return (SET_ERROR(EDQUOT));
1063 dsl_dir_rele(pdd, FTAG);
1065 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
1069 /* You can only clone snapshots, not the head datasets. */
1070 if (!origin->ds_is_snapshot) {
1071 dsl_dataset_rele(origin, FTAG);
1072 return (SET_ERROR(EINVAL));
1074 dsl_dataset_rele(origin, FTAG);
1080 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
1082 dmu_objset_clone_arg_t *doca = arg;
1083 dsl_pool_t *dp = dmu_tx_pool(tx);
1086 dsl_dataset_t *origin, *ds;
1088 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1090 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
1091 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
1093 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
1094 doca->doca_cred, tx);
1096 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
1097 dsl_dataset_name(origin, namebuf);
1098 spa_history_log_internal_ds(ds, "clone", tx,
1099 "origin=%s (%llu)", namebuf, origin->ds_object);
1100 dsl_dataset_rele(ds, FTAG);
1101 dsl_dataset_rele(origin, FTAG);
1102 dsl_dir_rele(pdd, FTAG);
1106 dmu_objset_clone(const char *clone, const char *origin)
1108 dmu_objset_clone_arg_t doca;
1110 doca.doca_clone = clone;
1111 doca.doca_origin = origin;
1112 doca.doca_cred = CRED();
1114 return (dsl_sync_task(clone,
1115 dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
1116 5, ZFS_SPACE_CHECK_NORMAL));
1120 dmu_objset_remap_indirects_impl(objset_t *os, uint64_t last_removed_txg)
1123 uint64_t object = 0;
1124 while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) {
1125 error = dmu_object_remap_indirects(os, object,
1128 * If the ZPL removed the object before we managed to dnode_hold
1129 * it, we would get an ENOENT. If the ZPL declares its intent
1130 * to remove the object (dnode_free) before we manage to
1131 * dnode_hold it, we would get an EEXIST. In either case, we
1132 * want to continue remapping the other objects in the objset;
1133 * in all other cases, we want to break early.
1135 if (error != 0 && error != ENOENT && error != EEXIST) {
1139 if (error == ESRCH) {
1146 dmu_objset_remap_indirects(const char *fsname)
1149 objset_t *os = NULL;
1150 uint64_t last_removed_txg;
1151 uint64_t remap_start_txg;
1154 error = dmu_objset_hold(fsname, FTAG, &os);
1158 dd = dmu_objset_ds(os)->ds_dir;
1160 if (!spa_feature_is_enabled(dmu_objset_spa(os),
1161 SPA_FEATURE_OBSOLETE_COUNTS)) {
1162 dmu_objset_rele(os, FTAG);
1163 return (SET_ERROR(ENOTSUP));
1166 if (dsl_dataset_is_snapshot(dmu_objset_ds(os))) {
1167 dmu_objset_rele(os, FTAG);
1168 return (SET_ERROR(EINVAL));
1172 * If there has not been a removal, we're done.
1174 last_removed_txg = spa_get_last_removal_txg(dmu_objset_spa(os));
1175 if (last_removed_txg == -1ULL) {
1176 dmu_objset_rele(os, FTAG);
1181 * If we have remapped since the last removal, we're done.
1183 if (dsl_dir_is_zapified(dd)) {
1184 uint64_t last_remap_txg;
1185 if (zap_lookup(spa_meta_objset(dmu_objset_spa(os)),
1186 dd->dd_object, DD_FIELD_LAST_REMAP_TXG,
1187 sizeof (last_remap_txg), 1, &last_remap_txg) == 0 &&
1188 last_remap_txg > last_removed_txg) {
1189 dmu_objset_rele(os, FTAG);
1194 dsl_dataset_long_hold(dmu_objset_ds(os), FTAG);
1195 dsl_pool_rele(dmu_objset_pool(os), FTAG);
1197 remap_start_txg = spa_last_synced_txg(dmu_objset_spa(os));
1198 error = dmu_objset_remap_indirects_impl(os, last_removed_txg);
1201 * We update the last_remap_txg to be the start txg so that
1202 * we can guarantee that every block older than last_remap_txg
1203 * that can be remapped has been remapped.
1205 error = dsl_dir_update_last_remap_txg(dd, remap_start_txg);
1208 dsl_dataset_long_rele(dmu_objset_ds(os), FTAG);
1209 dsl_dataset_rele(dmu_objset_ds(os), FTAG);
1215 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
1218 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1219 nvlist_t *snaps = fnvlist_alloc();
1221 fnvlist_add_boolean(snaps, longsnap);
1223 err = dsl_dataset_snapshot(snaps, NULL, NULL);
1224 fnvlist_free(snaps);
1229 dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx)
1233 while ((dn = multilist_sublist_head(list)) != NULL) {
1234 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1235 ASSERT(dn->dn_dbuf->db_data_pending);
1237 * Initialize dn_zio outside dnode_sync() because the
1238 * meta-dnode needs to set it ouside dnode_sync().
1240 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1243 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1244 multilist_sublist_remove(list, dn);
1246 multilist_t *newlist = dn->dn_objset->os_synced_dnodes;
1247 if (newlist != NULL) {
1248 (void) dnode_add_ref(dn, newlist);
1249 multilist_insert(newlist, dn);
1258 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1260 blkptr_t *bp = zio->io_bp;
1262 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1264 ASSERT(!BP_IS_EMBEDDED(bp));
1265 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1266 ASSERT0(BP_GET_LEVEL(bp));
1269 * Update rootbp fill count: it should be the number of objects
1270 * allocated in the object set (not counting the "special"
1271 * objects that are stored in the objset_phys_t -- the meta
1272 * dnode and user/group accounting objects).
1275 for (int i = 0; i < dnp->dn_nblkptr; i++)
1276 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1277 if (os->os_dsl_dataset != NULL)
1278 rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG);
1279 *os->os_rootbp = *bp;
1280 if (os->os_dsl_dataset != NULL)
1281 rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG);
1286 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1288 blkptr_t *bp = zio->io_bp;
1289 blkptr_t *bp_orig = &zio->io_bp_orig;
1292 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1293 ASSERT(BP_EQUAL(bp, bp_orig));
1295 dsl_dataset_t *ds = os->os_dsl_dataset;
1296 dmu_tx_t *tx = os->os_synctx;
1298 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1299 dsl_dataset_block_born(ds, bp, tx);
1301 kmem_free(bp, sizeof (*bp));
1304 typedef struct sync_dnodes_arg {
1305 multilist_t *sda_list;
1306 int sda_sublist_idx;
1307 multilist_t *sda_newlist;
1309 } sync_dnodes_arg_t;
1312 sync_dnodes_task(void *arg)
1314 sync_dnodes_arg_t *sda = arg;
1316 multilist_sublist_t *ms =
1317 multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx);
1319 dmu_objset_sync_dnodes(ms, sda->sda_tx);
1321 multilist_sublist_unlock(ms);
1323 kmem_free(sda, sizeof (*sda));
1327 /* called from dsl */
1329 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1332 zbookmark_phys_t zb;
1336 dbuf_dirty_record_t *dr;
1337 blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP);
1338 *blkptr_copy = *os->os_rootbp;
1340 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1342 ASSERT(dmu_tx_is_syncing(tx));
1343 /* XXX the write_done callback should really give us the tx... */
1346 if (os->os_dsl_dataset == NULL) {
1348 * This is the MOS. If we have upgraded,
1349 * spa_max_replication() could change, so reset
1352 os->os_copies = spa_max_replication(os->os_spa);
1356 * Create the root block IO
1358 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1359 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1360 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1361 arc_release(os->os_phys_buf, &os->os_phys_buf);
1363 dmu_write_policy(os, NULL, 0, 0, &zp);
1365 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1366 blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1367 &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done,
1368 os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
1371 * Sync special dnodes - the parent IO for the sync is the root block
1373 DMU_META_DNODE(os)->dn_zio = zio;
1374 dnode_sync(DMU_META_DNODE(os), tx);
1376 os->os_phys->os_flags = os->os_flags;
1378 if (DMU_USERUSED_DNODE(os) &&
1379 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1380 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1381 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1382 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1383 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1386 txgoff = tx->tx_txg & TXG_MASK;
1388 if (dmu_objset_userused_enabled(os)) {
1390 * We must create the list here because it uses the
1391 * dn_dirty_link[] of this txg. But it may already
1392 * exist because we call dsl_dataset_sync() twice per txg.
1394 if (os->os_synced_dnodes == NULL) {
1395 os->os_synced_dnodes =
1396 multilist_create(sizeof (dnode_t),
1397 offsetof(dnode_t, dn_dirty_link[txgoff]),
1398 dnode_multilist_index_func);
1400 ASSERT3U(os->os_synced_dnodes->ml_offset, ==,
1401 offsetof(dnode_t, dn_dirty_link[txgoff]));
1406 i < multilist_get_num_sublists(os->os_dirty_dnodes[txgoff]); i++) {
1407 sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP);
1408 sda->sda_list = os->os_dirty_dnodes[txgoff];
1409 sda->sda_sublist_idx = i;
1411 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
1412 sync_dnodes_task, sda, 0);
1413 /* callback frees sda */
1415 taskq_wait(dmu_objset_pool(os)->dp_sync_taskq);
1417 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1418 while ((dr = list_head(list)) != NULL) {
1419 ASSERT0(dr->dr_dbuf->db_level);
1420 list_remove(list, dr);
1422 zio_nowait(dr->dr_zio);
1425 /* Enable dnode backfill if enough objects have been freed. */
1426 if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) {
1427 os->os_rescan_dnodes = B_TRUE;
1428 os->os_freed_dnodes = 0;
1432 * Free intent log blocks up to this tx.
1434 zil_sync(os->os_zil, tx);
1435 os->os_phys->os_zil_header = os->os_zil_header;
1440 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1442 return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK]));
1445 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1448 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1454 dmu_objset_userused_enabled(objset_t *os)
1456 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1457 used_cbs[os->os_phys->os_type] != NULL &&
1458 DMU_USERUSED_DNODE(os) != NULL);
1461 typedef struct userquota_node {
1464 avl_node_t uqn_node;
1467 typedef struct userquota_cache {
1468 avl_tree_t uqc_user_deltas;
1469 avl_tree_t uqc_group_deltas;
1470 } userquota_cache_t;
1473 userquota_compare(const void *l, const void *r)
1475 const userquota_node_t *luqn = l;
1476 const userquota_node_t *ruqn = r;
1478 if (luqn->uqn_id < ruqn->uqn_id)
1480 if (luqn->uqn_id > ruqn->uqn_id)
1486 do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx)
1489 userquota_node_t *uqn;
1491 ASSERT(dmu_tx_is_syncing(tx));
1494 while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas,
1495 &cookie)) != NULL) {
1497 * os_userused_lock protects against concurrent calls to
1498 * zap_increment_int(). It's needed because zap_increment_int()
1499 * is not thread-safe (i.e. not atomic).
1501 mutex_enter(&os->os_userused_lock);
1502 VERIFY0(zap_increment_int(os, DMU_USERUSED_OBJECT,
1503 uqn->uqn_id, uqn->uqn_delta, tx));
1504 mutex_exit(&os->os_userused_lock);
1505 kmem_free(uqn, sizeof (*uqn));
1507 avl_destroy(&cache->uqc_user_deltas);
1510 while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas,
1511 &cookie)) != NULL) {
1512 mutex_enter(&os->os_userused_lock);
1513 VERIFY0(zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1514 uqn->uqn_id, uqn->uqn_delta, tx));
1515 mutex_exit(&os->os_userused_lock);
1516 kmem_free(uqn, sizeof (*uqn));
1518 avl_destroy(&cache->uqc_group_deltas);
1522 userquota_update_cache(avl_tree_t *avl, uint64_t id, int64_t delta)
1524 userquota_node_t search = { .uqn_id = id };
1527 userquota_node_t *uqn = avl_find(avl, &search, &idx);
1529 uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP);
1531 avl_insert(avl, uqn, idx);
1533 uqn->uqn_delta += delta;
1537 do_userquota_update(userquota_cache_t *cache, uint64_t used, uint64_t flags,
1538 uint64_t user, uint64_t group, boolean_t subtract)
1540 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1541 int64_t delta = DNODE_MIN_SIZE + used;
1545 userquota_update_cache(&cache->uqc_user_deltas, user, delta);
1546 userquota_update_cache(&cache->uqc_group_deltas, group, delta);
1550 typedef struct userquota_updates_arg {
1552 int uua_sublist_idx;
1554 } userquota_updates_arg_t;
1557 userquota_updates_task(void *arg)
1559 userquota_updates_arg_t *uua = arg;
1560 objset_t *os = uua->uua_os;
1561 dmu_tx_t *tx = uua->uua_tx;
1563 userquota_cache_t cache = { 0 };
1565 multilist_sublist_t *list =
1566 multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx);
1568 ASSERT(multilist_sublist_head(list) == NULL ||
1569 dmu_objset_userused_enabled(os));
1570 avl_create(&cache.uqc_user_deltas, userquota_compare,
1571 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1572 avl_create(&cache.uqc_group_deltas, userquota_compare,
1573 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1575 while ((dn = multilist_sublist_head(list)) != NULL) {
1577 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1578 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1579 dn->dn_phys->dn_flags &
1580 DNODE_FLAG_USERUSED_ACCOUNTED);
1582 flags = dn->dn_id_flags;
1584 if (flags & DN_ID_OLD_EXIST) {
1585 do_userquota_update(&cache,
1586 dn->dn_oldused, dn->dn_oldflags,
1587 dn->dn_olduid, dn->dn_oldgid, B_TRUE);
1589 if (flags & DN_ID_NEW_EXIST) {
1590 do_userquota_update(&cache,
1591 DN_USED_BYTES(dn->dn_phys),
1592 dn->dn_phys->dn_flags, dn->dn_newuid,
1593 dn->dn_newgid, B_FALSE);
1596 mutex_enter(&dn->dn_mtx);
1598 dn->dn_oldflags = 0;
1599 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1600 dn->dn_olduid = dn->dn_newuid;
1601 dn->dn_oldgid = dn->dn_newgid;
1602 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1603 if (dn->dn_bonuslen == 0)
1604 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1606 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1608 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1609 mutex_exit(&dn->dn_mtx);
1611 multilist_sublist_remove(list, dn);
1612 dnode_rele(dn, os->os_synced_dnodes);
1614 do_userquota_cacheflush(os, &cache, tx);
1615 multilist_sublist_unlock(list);
1616 kmem_free(uua, sizeof (*uua));
1620 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1622 if (!dmu_objset_userused_enabled(os))
1625 /* Allocate the user/groupused objects if necessary. */
1626 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1627 VERIFY0(zap_create_claim(os,
1628 DMU_USERUSED_OBJECT,
1629 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1630 VERIFY0(zap_create_claim(os,
1631 DMU_GROUPUSED_OBJECT,
1632 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1636 i < multilist_get_num_sublists(os->os_synced_dnodes); i++) {
1637 userquota_updates_arg_t *uua =
1638 kmem_alloc(sizeof (*uua), KM_SLEEP);
1640 uua->uua_sublist_idx = i;
1642 /* note: caller does taskq_wait() */
1643 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
1644 userquota_updates_task, uua, 0);
1645 /* callback frees uua */
1650 * Returns a pointer to data to find uid/gid from
1652 * If a dirty record for transaction group that is syncing can't
1653 * be found then NULL is returned. In the NULL case it is assumed
1654 * the uid/gid aren't changing.
1657 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1659 dbuf_dirty_record_t *dr, **drp;
1662 if (db->db_dirtycnt == 0)
1663 return (db->db.db_data); /* Nothing is changing */
1665 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1666 if (dr->dr_txg == tx->tx_txg)
1674 DB_DNODE_ENTER(dr->dr_dbuf);
1675 dn = DB_DNODE(dr->dr_dbuf);
1677 if (dn->dn_bonuslen == 0 &&
1678 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1679 data = dr->dt.dl.dr_data->b_data;
1681 data = dr->dt.dl.dr_data;
1683 DB_DNODE_EXIT(dr->dr_dbuf);
1690 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1692 objset_t *os = dn->dn_objset;
1694 dmu_buf_impl_t *db = NULL;
1695 uint64_t *user = NULL;
1696 uint64_t *group = NULL;
1697 int flags = dn->dn_id_flags;
1699 boolean_t have_spill = B_FALSE;
1701 if (!dmu_objset_userused_enabled(dn->dn_objset))
1704 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1705 DN_ID_CHKED_SPILL)))
1708 if (before && dn->dn_bonuslen != 0)
1709 data = DN_BONUS(dn->dn_phys);
1710 else if (!before && dn->dn_bonuslen != 0) {
1713 mutex_enter(&db->db_mtx);
1714 data = dmu_objset_userquota_find_data(db, tx);
1716 data = DN_BONUS(dn->dn_phys);
1718 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1721 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1722 rf |= DB_RF_HAVESTRUCT;
1723 error = dmu_spill_hold_by_dnode(dn,
1724 rf | DB_RF_MUST_SUCCEED,
1725 FTAG, (dmu_buf_t **)&db);
1727 mutex_enter(&db->db_mtx);
1728 data = (before) ? db->db.db_data :
1729 dmu_objset_userquota_find_data(db, tx);
1730 have_spill = B_TRUE;
1732 mutex_enter(&dn->dn_mtx);
1733 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1734 mutex_exit(&dn->dn_mtx);
1740 user = &dn->dn_olduid;
1741 group = &dn->dn_oldgid;
1743 user = &dn->dn_newuid;
1744 group = &dn->dn_newgid;
1748 * Must always call the callback in case the object
1749 * type has changed and that type isn't an object type to track
1751 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1755 * Preserve existing uid/gid when the callback can't determine
1756 * what the new uid/gid are and the callback returned EEXIST.
1757 * The EEXIST error tells us to just use the existing uid/gid.
1758 * If we don't know what the old values are then just assign
1759 * them to 0, since that is a new file being created.
1761 if (!before && data == NULL && error == EEXIST) {
1762 if (flags & DN_ID_OLD_EXIST) {
1763 dn->dn_newuid = dn->dn_olduid;
1764 dn->dn_newgid = dn->dn_oldgid;
1773 mutex_exit(&db->db_mtx);
1775 mutex_enter(&dn->dn_mtx);
1776 if (error == 0 && before)
1777 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1778 if (error == 0 && !before)
1779 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1782 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1784 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1786 mutex_exit(&dn->dn_mtx);
1788 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1792 dmu_objset_userspace_present(objset_t *os)
1794 return (os->os_phys->os_flags &
1795 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1799 dmu_objset_userspace_upgrade(objset_t *os)
1804 if (dmu_objset_userspace_present(os))
1806 if (!dmu_objset_userused_enabled(os))
1807 return (SET_ERROR(ENOTSUP));
1808 if (dmu_objset_is_snapshot(os))
1809 return (SET_ERROR(EINVAL));
1812 * We simply need to mark every object dirty, so that it will be
1813 * synced out and now accounted. If this is called
1814 * concurrently, or if we already did some work before crashing,
1815 * that's fine, since we track each object's accounted state
1819 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1824 if (issig(JUSTLOOKING) && issig(FORREAL))
1825 return (SET_ERROR(EINTR));
1827 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1830 tx = dmu_tx_create(os);
1831 dmu_tx_hold_bonus(tx, obj);
1832 objerr = dmu_tx_assign(tx, TXG_WAIT);
1837 dmu_buf_will_dirty(db, tx);
1838 dmu_buf_rele(db, FTAG);
1842 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1843 txg_wait_synced(dmu_objset_pool(os), 0);
1848 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1849 uint64_t *usedobjsp, uint64_t *availobjsp)
1851 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1852 usedobjsp, availobjsp);
1856 dmu_objset_fsid_guid(objset_t *os)
1858 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1862 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1864 stat->dds_type = os->os_phys->os_type;
1865 if (os->os_dsl_dataset)
1866 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1870 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1872 ASSERT(os->os_dsl_dataset ||
1873 os->os_phys->os_type == DMU_OST_META);
1875 if (os->os_dsl_dataset != NULL)
1876 dsl_dataset_stats(os->os_dsl_dataset, nv);
1878 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1879 os->os_phys->os_type);
1880 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1881 dmu_objset_userspace_present(os));
1885 dmu_objset_is_snapshot(objset_t *os)
1887 if (os->os_dsl_dataset != NULL)
1888 return (os->os_dsl_dataset->ds_is_snapshot);
1894 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1895 boolean_t *conflict)
1897 dsl_dataset_t *ds = os->os_dsl_dataset;
1900 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1901 return (SET_ERROR(ENOENT));
1903 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1904 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
1905 MT_NORMALIZE, real, maxlen, conflict));
1909 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1910 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1912 dsl_dataset_t *ds = os->os_dsl_dataset;
1913 zap_cursor_t cursor;
1914 zap_attribute_t attr;
1916 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1918 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1919 return (SET_ERROR(ENOENT));
1921 zap_cursor_init_serialized(&cursor,
1922 ds->ds_dir->dd_pool->dp_meta_objset,
1923 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
1925 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1926 zap_cursor_fini(&cursor);
1927 return (SET_ERROR(ENOENT));
1930 if (strlen(attr.za_name) + 1 > namelen) {
1931 zap_cursor_fini(&cursor);
1932 return (SET_ERROR(ENAMETOOLONG));
1935 (void) strcpy(name, attr.za_name);
1937 *idp = attr.za_first_integer;
1939 *case_conflict = attr.za_normalization_conflict;
1940 zap_cursor_advance(&cursor);
1941 *offp = zap_cursor_serialize(&cursor);
1942 zap_cursor_fini(&cursor);
1948 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1949 uint64_t *idp, uint64_t *offp)
1951 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1952 zap_cursor_t cursor;
1953 zap_attribute_t attr;
1955 /* there is no next dir on a snapshot! */
1956 if (os->os_dsl_dataset->ds_object !=
1957 dsl_dir_phys(dd)->dd_head_dataset_obj)
1958 return (SET_ERROR(ENOENT));
1960 zap_cursor_init_serialized(&cursor,
1961 dd->dd_pool->dp_meta_objset,
1962 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
1964 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1965 zap_cursor_fini(&cursor);
1966 return (SET_ERROR(ENOENT));
1969 if (strlen(attr.za_name) + 1 > namelen) {
1970 zap_cursor_fini(&cursor);
1971 return (SET_ERROR(ENAMETOOLONG));
1974 (void) strcpy(name, attr.za_name);
1976 *idp = attr.za_first_integer;
1977 zap_cursor_advance(&cursor);
1978 *offp = zap_cursor_serialize(&cursor);
1979 zap_cursor_fini(&cursor);
1984 typedef struct dmu_objset_find_ctx {
1988 char *dc_ddname; /* last component of ddobj's name */
1989 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *);
1992 kmutex_t *dc_error_lock;
1994 } dmu_objset_find_ctx_t;
1997 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp)
1999 dsl_pool_t *dp = dcp->dc_dp;
2003 zap_attribute_t *attr;
2007 /* don't process if there already was an error */
2008 if (*dcp->dc_error != 0)
2012 * Note: passing the name (dc_ddname) here is optional, but it
2013 * improves performance because we don't need to call
2014 * zap_value_search() to determine the name.
2016 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd);
2020 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2021 if (dd->dd_myname[0] == '$') {
2022 dsl_dir_rele(dd, FTAG);
2026 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
2027 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
2030 * Iterate over all children.
2032 if (dcp->dc_flags & DS_FIND_CHILDREN) {
2033 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2034 dsl_dir_phys(dd)->dd_child_dir_zapobj);
2035 zap_cursor_retrieve(&zc, attr) == 0;
2036 (void) zap_cursor_advance(&zc)) {
2037 ASSERT3U(attr->za_integer_length, ==,
2039 ASSERT3U(attr->za_num_integers, ==, 1);
2041 dmu_objset_find_ctx_t *child_dcp =
2042 kmem_alloc(sizeof (*child_dcp), KM_SLEEP);
2044 child_dcp->dc_ddobj = attr->za_first_integer;
2045 child_dcp->dc_ddname = spa_strdup(attr->za_name);
2046 if (dcp->dc_tq != NULL)
2047 (void) taskq_dispatch(dcp->dc_tq,
2048 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP);
2050 dmu_objset_find_dp_impl(child_dcp);
2052 zap_cursor_fini(&zc);
2056 * Iterate over all snapshots.
2058 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) {
2060 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2065 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2066 dsl_dataset_rele(ds, FTAG);
2068 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2069 zap_cursor_retrieve(&zc, attr) == 0;
2070 (void) zap_cursor_advance(&zc)) {
2071 ASSERT3U(attr->za_integer_length, ==,
2073 ASSERT3U(attr->za_num_integers, ==, 1);
2075 err = dsl_dataset_hold_obj(dp,
2076 attr->za_first_integer, FTAG, &ds);
2079 err = dcp->dc_func(dp, ds, dcp->dc_arg);
2080 dsl_dataset_rele(ds, FTAG);
2084 zap_cursor_fini(&zc);
2088 kmem_free(attr, sizeof (zap_attribute_t));
2091 dsl_dir_rele(dd, FTAG);
2098 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2101 * Note: we hold the dir while calling dsl_dataset_hold_obj() so
2102 * that the dir will remain cached, and we won't have to re-instantiate
2103 * it (which could be expensive due to finding its name via
2104 * zap_value_search()).
2106 dsl_dir_rele(dd, FTAG);
2109 err = dcp->dc_func(dp, ds, dcp->dc_arg);
2110 dsl_dataset_rele(ds, FTAG);
2114 mutex_enter(dcp->dc_error_lock);
2115 /* only keep first error */
2116 if (*dcp->dc_error == 0)
2117 *dcp->dc_error = err;
2118 mutex_exit(dcp->dc_error_lock);
2121 if (dcp->dc_ddname != NULL)
2122 spa_strfree(dcp->dc_ddname);
2123 kmem_free(dcp, sizeof (*dcp));
2127 dmu_objset_find_dp_cb(void *arg)
2129 dmu_objset_find_ctx_t *dcp = arg;
2130 dsl_pool_t *dp = dcp->dc_dp;
2133 * We need to get a pool_config_lock here, as there are several
2134 * asssert(pool_config_held) down the stack. Getting a lock via
2135 * dsl_pool_config_enter is risky, as it might be stalled by a
2136 * pending writer. This would deadlock, as the write lock can
2137 * only be granted when our parent thread gives up the lock.
2138 * The _prio interface gives us priority over a pending writer.
2140 dsl_pool_config_enter_prio(dp, FTAG);
2142 dmu_objset_find_dp_impl(dcp);
2144 dsl_pool_config_exit(dp, FTAG);
2148 * Find objsets under and including ddobj, call func(ds) on each.
2149 * The order for the enumeration is completely undefined.
2150 * func is called with dsl_pool_config held.
2153 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
2154 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
2159 dmu_objset_find_ctx_t *dcp;
2162 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL);
2163 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP);
2166 dcp->dc_ddobj = ddobj;
2167 dcp->dc_ddname = NULL;
2168 dcp->dc_func = func;
2170 dcp->dc_flags = flags;
2171 dcp->dc_error_lock = &err_lock;
2172 dcp->dc_error = &error;
2174 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) {
2176 * In case a write lock is held we can't make use of
2177 * parallelism, as down the stack of the worker threads
2178 * the lock is asserted via dsl_pool_config_held.
2179 * In case of a read lock this is solved by getting a read
2180 * lock in each worker thread, which isn't possible in case
2181 * of a writer lock. So we fall back to the synchronous path
2183 * In the future it might be possible to get some magic into
2184 * dsl_pool_config_held in a way that it returns true for
2185 * the worker threads so that a single lock held from this
2186 * thread suffices. For now, stay single threaded.
2188 dmu_objset_find_dp_impl(dcp);
2189 mutex_destroy(&err_lock);
2194 ntasks = dmu_find_threads;
2196 ntasks = vdev_count_leaves(dp->dp_spa) * 4;
2197 tq = taskq_create("dmu_objset_find", ntasks, minclsyspri, ntasks,
2200 kmem_free(dcp, sizeof (*dcp));
2201 mutex_destroy(&err_lock);
2203 return (SET_ERROR(ENOMEM));
2207 /* dcp will be freed by task */
2208 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP);
2211 * PORTING: this code relies on the property of taskq_wait to wait
2212 * until no more tasks are queued and no more tasks are active. As
2213 * we always queue new tasks from within other tasks, task_wait
2214 * reliably waits for the full recursion to finish, even though we
2215 * enqueue new tasks after taskq_wait has been called.
2216 * On platforms other than illumos, taskq_wait may not have this
2221 mutex_destroy(&err_lock);
2227 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
2228 * The dp_config_rwlock must not be held when this is called, and it
2229 * will not be held when the callback is called.
2230 * Therefore this function should only be used when the pool is not changing
2231 * (e.g. in syncing context), or the callback can deal with the possible races.
2234 dmu_objset_find_impl(spa_t *spa, const char *name,
2235 int func(const char *, void *), void *arg, int flags)
2238 dsl_pool_t *dp = spa_get_dsl(spa);
2241 zap_attribute_t *attr;
2246 dsl_pool_config_enter(dp, FTAG);
2248 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
2250 dsl_pool_config_exit(dp, FTAG);
2254 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2255 if (dd->dd_myname[0] == '$') {
2256 dsl_dir_rele(dd, FTAG);
2257 dsl_pool_config_exit(dp, FTAG);
2261 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
2262 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
2265 * Iterate over all children.
2267 if (flags & DS_FIND_CHILDREN) {
2268 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2269 dsl_dir_phys(dd)->dd_child_dir_zapobj);
2270 zap_cursor_retrieve(&zc, attr) == 0;
2271 (void) zap_cursor_advance(&zc)) {
2272 ASSERT3U(attr->za_integer_length, ==,
2274 ASSERT3U(attr->za_num_integers, ==, 1);
2276 child = kmem_asprintf("%s/%s", name, attr->za_name);
2277 dsl_pool_config_exit(dp, FTAG);
2278 err = dmu_objset_find_impl(spa, child,
2280 dsl_pool_config_enter(dp, FTAG);
2285 zap_cursor_fini(&zc);
2288 dsl_dir_rele(dd, FTAG);
2289 dsl_pool_config_exit(dp, FTAG);
2290 kmem_free(attr, sizeof (zap_attribute_t));
2296 * Iterate over all snapshots.
2298 if (flags & DS_FIND_SNAPSHOTS) {
2299 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2304 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2305 dsl_dataset_rele(ds, FTAG);
2307 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2308 zap_cursor_retrieve(&zc, attr) == 0;
2309 (void) zap_cursor_advance(&zc)) {
2310 ASSERT3U(attr->za_integer_length, ==,
2312 ASSERT3U(attr->za_num_integers, ==, 1);
2314 child = kmem_asprintf("%s@%s",
2315 name, attr->za_name);
2316 dsl_pool_config_exit(dp, FTAG);
2317 err = func(child, arg);
2318 dsl_pool_config_enter(dp, FTAG);
2323 zap_cursor_fini(&zc);
2327 dsl_dir_rele(dd, FTAG);
2328 kmem_free(attr, sizeof (zap_attribute_t));
2329 dsl_pool_config_exit(dp, FTAG);
2334 /* Apply to self. */
2335 return (func(name, arg));
2339 * See comment above dmu_objset_find_impl().
2342 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
2348 error = spa_open(name, &spa, FTAG);
2351 error = dmu_objset_find_impl(spa, name, func, arg, flags);
2352 spa_close(spa, FTAG);
2357 dmu_objset_set_user(objset_t *os, void *user_ptr)
2359 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2360 os->os_user_ptr = user_ptr;
2364 dmu_objset_get_user(objset_t *os)
2366 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2367 return (os->os_user_ptr);
2371 * Determine name of filesystem, given name of snapshot.
2372 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
2375 dmu_fsname(const char *snapname, char *buf)
2377 char *atp = strchr(snapname, '@');
2379 return (SET_ERROR(EINVAL));
2380 if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN)
2381 return (SET_ERROR(ENAMETOOLONG));
2382 (void) strlcpy(buf, snapname, atp - snapname + 1);
2387 * Call when we think we're going to write/free space in open context to track
2388 * the amount of dirty data in the open txg, which is also the amount
2389 * of memory that can not be evicted until this txg syncs.
2392 dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx)
2394 dsl_dataset_t *ds = os->os_dsl_dataset;
2395 int64_t aspace = spa_get_worst_case_asize(os->os_spa, space);
2398 dsl_dir_willuse_space(ds->ds_dir, aspace, tx);
2399 dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx);