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) 2016 Actifio, Inc. All rights reserved.
30 * Copyright 2017 Nexenta Systems, Inc.
33 /* Portions Copyright 2010 Robert Milkowski */
35 #include <sys/zfeature.h>
37 #include <sys/zfs_context.h>
38 #include <sys/dmu_objset.h>
39 #include <sys/dsl_dir.h>
40 #include <sys/dsl_dataset.h>
41 #include <sys/dsl_prop.h>
42 #include <sys/dsl_pool.h>
43 #include <sys/dsl_synctask.h>
44 #include <sys/dsl_deleg.h>
45 #include <sys/dnode.h>
48 #include <sys/dmu_tx.h>
51 #include <sys/dmu_impl.h>
52 #include <sys/zfs_ioctl.h>
54 #include <sys/zfs_onexit.h>
55 #include <sys/dsl_destroy.h>
57 #include <sys/policy.h>
58 #include <sys/spa_impl.h>
61 * Needed to close a window in dnode_move() that allows the objset to be freed
62 * before it can be safely accessed.
67 * Tunable to overwrite the maximum number of threads for the parallelization
68 * of dmu_objset_find_dp, needed to speed up the import of pools with many
70 * Default is 4 times the number of leaf vdevs.
72 int dmu_find_threads = 0;
75 * Backfill lower metadnode objects after this many have been freed.
76 * Backfilling negatively impacts object creation rates, so only do it
77 * if there are enough holes to fill.
79 int dmu_rescan_dnode_threshold = 1 << DN_MAX_INDBLKSHIFT;
81 static void dmu_objset_find_dp_cb(void *arg);
83 static void dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb);
84 static void dmu_objset_upgrade_stop(objset_t *os);
89 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
99 dmu_objset_spa(objset_t *os)
105 dmu_objset_zil(objset_t *os)
111 dmu_objset_pool(objset_t *os)
115 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
116 return (ds->ds_dir->dd_pool);
118 return (spa_get_dsl(os->os_spa));
122 dmu_objset_ds(objset_t *os)
124 return (os->os_dsl_dataset);
128 dmu_objset_type(objset_t *os)
130 return (os->os_phys->os_type);
134 dmu_objset_name(objset_t *os, char *buf)
136 dsl_dataset_name(os->os_dsl_dataset, buf);
140 dmu_objset_id(objset_t *os)
142 dsl_dataset_t *ds = os->os_dsl_dataset;
144 return (ds ? ds->ds_object : 0);
148 dmu_objset_dnodesize(objset_t *os)
150 return (os->os_dnodesize);
154 dmu_objset_syncprop(objset_t *os)
156 return (os->os_sync);
160 dmu_objset_logbias(objset_t *os)
162 return (os->os_logbias);
166 checksum_changed_cb(void *arg, uint64_t newval)
171 * Inheritance should have been done by now.
173 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
175 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
179 compression_changed_cb(void *arg, uint64_t newval)
184 * Inheritance and range checking should have been done by now.
186 ASSERT(newval != ZIO_COMPRESS_INHERIT);
188 os->os_compress = zio_compress_select(os->os_spa, newval,
193 copies_changed_cb(void *arg, uint64_t newval)
198 * Inheritance and range checking should have been done by now.
201 ASSERT(newval <= spa_max_replication(os->os_spa));
203 os->os_copies = newval;
207 dedup_changed_cb(void *arg, uint64_t newval)
210 spa_t *spa = os->os_spa;
211 enum zio_checksum checksum;
214 * Inheritance should have been done by now.
216 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
218 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
220 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
221 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
225 primary_cache_changed_cb(void *arg, uint64_t newval)
230 * Inheritance and range checking should have been done by now.
232 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
233 newval == ZFS_CACHE_METADATA);
235 os->os_primary_cache = newval;
239 secondary_cache_changed_cb(void *arg, uint64_t newval)
244 * Inheritance and range checking should have been done by now.
246 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
247 newval == ZFS_CACHE_METADATA);
249 os->os_secondary_cache = newval;
253 sync_changed_cb(void *arg, uint64_t newval)
258 * Inheritance and range checking should have been done by now.
260 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
261 newval == ZFS_SYNC_DISABLED);
263 os->os_sync = newval;
265 zil_set_sync(os->os_zil, newval);
269 redundant_metadata_changed_cb(void *arg, uint64_t newval)
274 * Inheritance and range checking should have been done by now.
276 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
277 newval == ZFS_REDUNDANT_METADATA_MOST);
279 os->os_redundant_metadata = newval;
283 dnodesize_changed_cb(void *arg, uint64_t newval)
288 case ZFS_DNSIZE_LEGACY:
289 os->os_dnodesize = DNODE_MIN_SIZE;
291 case ZFS_DNSIZE_AUTO:
293 * Choose a dnode size that will work well for most
294 * workloads if the user specified "auto". Future code
295 * improvements could dynamically select a dnode size
296 * based on observed workload patterns.
298 os->os_dnodesize = DNODE_MIN_SIZE * 2;
305 os->os_dnodesize = newval;
311 logbias_changed_cb(void *arg, uint64_t newval)
315 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
316 newval == ZFS_LOGBIAS_THROUGHPUT);
317 os->os_logbias = newval;
319 zil_set_logbias(os->os_zil, newval);
323 recordsize_changed_cb(void *arg, uint64_t newval)
327 os->os_recordsize = newval;
331 dmu_objset_byteswap(void *buf, size_t size)
333 objset_phys_t *osp = buf;
335 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
336 dnode_byteswap(&osp->os_meta_dnode);
337 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
338 osp->os_type = BSWAP_64(osp->os_type);
339 osp->os_flags = BSWAP_64(osp->os_flags);
340 if (size == sizeof (objset_phys_t)) {
341 dnode_byteswap(&osp->os_userused_dnode);
342 dnode_byteswap(&osp->os_groupused_dnode);
347 * The hash is a CRC-based hash of the objset_t pointer and the object number.
350 dnode_hash(const objset_t *os, uint64_t obj)
352 uintptr_t osv = (uintptr_t)os;
353 uint64_t crc = -1ULL;
355 ASSERT(zfs_crc64_table[128] == ZFS_CRC64_POLY);
357 * The low 6 bits of the pointer don't have much entropy, because
358 * the objset_t is larger than 2^6 bytes long.
360 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (osv >> 6)) & 0xFF];
361 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 0)) & 0xFF];
362 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 8)) & 0xFF];
363 crc = (crc >> 8) ^ zfs_crc64_table[(crc ^ (obj >> 16)) & 0xFF];
365 crc ^= (osv>>14) ^ (obj>>24);
371 dnode_multilist_index_func(multilist_t *ml, void *obj)
374 return (dnode_hash(dn->dn_objset, dn->dn_object) %
375 multilist_get_num_sublists(ml));
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));
387 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
388 os->os_dsl_dataset = ds;
391 if (!BP_IS_HOLE(os->os_rootbp)) {
392 arc_flags_t aflags = ARC_FLAG_WAIT;
394 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
395 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
397 if (DMU_OS_IS_L2CACHEABLE(os))
398 aflags |= ARC_FLAG_L2CACHE;
400 dprintf_bp(os->os_rootbp, "reading %s", "");
401 err = arc_read(NULL, spa, os->os_rootbp,
402 arc_getbuf_func, &os->os_phys_buf,
403 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
405 kmem_free(os, sizeof (objset_t));
406 /* convert checksum errors into IO errors */
408 err = SET_ERROR(EIO);
412 /* Increase the blocksize if we are permitted. */
413 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
414 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
415 arc_buf_t *buf = arc_alloc_buf(spa, &os->os_phys_buf,
416 ARC_BUFC_METADATA, sizeof (objset_phys_t));
417 bzero(buf->b_data, sizeof (objset_phys_t));
418 bcopy(os->os_phys_buf->b_data, buf->b_data,
419 arc_buf_size(os->os_phys_buf));
420 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
421 os->os_phys_buf = buf;
424 os->os_phys = os->os_phys_buf->b_data;
425 os->os_flags = os->os_phys->os_flags;
427 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
428 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
429 os->os_phys_buf = arc_alloc_buf(spa, &os->os_phys_buf,
430 ARC_BUFC_METADATA, size);
431 os->os_phys = os->os_phys_buf->b_data;
432 bzero(os->os_phys, size);
436 * Note: the changed_cb will be called once before the register
437 * func returns, thus changing the checksum/compression from the
438 * default (fletcher2/off). Snapshots don't need to know about
439 * checksum/compression/copies.
442 boolean_t needlock = B_FALSE;
445 * Note: it's valid to open the objset if the dataset is
446 * long-held, in which case the pool_config lock will not
449 if (!dsl_pool_config_held(dmu_objset_pool(os))) {
451 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
453 err = dsl_prop_register(ds,
454 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
455 primary_cache_changed_cb, os);
457 err = dsl_prop_register(ds,
458 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
459 secondary_cache_changed_cb, os);
461 if (!ds->ds_is_snapshot) {
463 err = dsl_prop_register(ds,
464 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
465 checksum_changed_cb, os);
468 err = dsl_prop_register(ds,
469 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
470 compression_changed_cb, os);
473 err = dsl_prop_register(ds,
474 zfs_prop_to_name(ZFS_PROP_COPIES),
475 copies_changed_cb, os);
478 err = dsl_prop_register(ds,
479 zfs_prop_to_name(ZFS_PROP_DEDUP),
480 dedup_changed_cb, os);
483 err = dsl_prop_register(ds,
484 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
485 logbias_changed_cb, os);
488 err = dsl_prop_register(ds,
489 zfs_prop_to_name(ZFS_PROP_SYNC),
490 sync_changed_cb, os);
493 err = dsl_prop_register(ds,
495 ZFS_PROP_REDUNDANT_METADATA),
496 redundant_metadata_changed_cb, os);
499 err = dsl_prop_register(ds,
500 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
501 recordsize_changed_cb, os);
504 err = dsl_prop_register(ds,
505 zfs_prop_to_name(ZFS_PROP_DNODESIZE),
506 dnodesize_changed_cb, os);
510 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
512 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
513 kmem_free(os, sizeof (objset_t));
517 /* It's the meta-objset. */
518 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
519 os->os_compress = ZIO_COMPRESS_ON;
520 os->os_copies = spa_max_replication(spa);
521 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
522 os->os_dedup_verify = B_FALSE;
523 os->os_logbias = ZFS_LOGBIAS_LATENCY;
524 os->os_sync = ZFS_SYNC_STANDARD;
525 os->os_primary_cache = ZFS_CACHE_ALL;
526 os->os_secondary_cache = ZFS_CACHE_ALL;
527 os->os_dnodesize = DNODE_MIN_SIZE;
530 if (ds == NULL || !ds->ds_is_snapshot)
531 os->os_zil_header = os->os_phys->os_zil_header;
532 os->os_zil = zil_alloc(os, &os->os_zil_header);
534 for (i = 0; i < TXG_SIZE; i++) {
535 os->os_dirty_dnodes[i] = multilist_create(sizeof (dnode_t),
536 offsetof(dnode_t, dn_dirty_link[i]),
537 dnode_multilist_index_func);
539 list_create(&os->os_dnodes, sizeof (dnode_t),
540 offsetof(dnode_t, dn_link));
541 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
542 offsetof(dmu_buf_impl_t, db_link));
544 list_link_init(&os->os_evicting_node);
546 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
547 mutex_init(&os->os_userused_lock, NULL, MUTEX_DEFAULT, NULL);
548 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
549 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
551 dnode_special_open(os, &os->os_phys->os_meta_dnode,
552 DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
553 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
554 dnode_special_open(os, &os->os_phys->os_userused_dnode,
555 DMU_USERUSED_OBJECT, &os->os_userused_dnode);
556 dnode_special_open(os, &os->os_phys->os_groupused_dnode,
557 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
560 mutex_init(&os->os_upgrade_lock, NULL, MUTEX_DEFAULT, NULL);
567 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
572 * We shouldn't be doing anything with dsl_dataset_t's unless the
573 * pool_config lock is held, or the dataset is long-held.
575 ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) ||
576 dsl_dataset_long_held(ds));
578 mutex_enter(&ds->ds_opening_lock);
579 if (ds->ds_objset == NULL) {
581 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
582 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
583 ds, dsl_dataset_get_blkptr(ds), &os);
584 rrw_exit(&ds->ds_bp_rwlock, FTAG);
587 mutex_enter(&ds->ds_lock);
588 ASSERT(ds->ds_objset == NULL);
590 mutex_exit(&ds->ds_lock);
593 *osp = ds->ds_objset;
594 mutex_exit(&ds->ds_opening_lock);
599 * Holds the pool while the objset is held. Therefore only one objset
600 * can be held at a time.
603 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
609 err = dsl_pool_hold(name, tag, &dp);
612 err = dsl_dataset_hold(dp, name, tag, &ds);
614 dsl_pool_rele(dp, tag);
618 err = dmu_objset_from_ds(ds, osp);
620 dsl_dataset_rele(ds, tag);
621 dsl_pool_rele(dp, tag);
628 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type,
629 boolean_t readonly, void *tag, objset_t **osp)
633 err = dmu_objset_from_ds(ds, osp);
635 dsl_dataset_disown(ds, tag);
636 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
637 dsl_dataset_disown(ds, tag);
638 return (SET_ERROR(EINVAL));
639 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
640 dsl_dataset_disown(ds, tag);
641 return (SET_ERROR(EROFS));
647 * dsl_pool must not be held when this is called.
648 * Upon successful return, there will be a longhold on the dataset,
649 * and the dsl_pool will not be held.
652 dmu_objset_own(const char *name, dmu_objset_type_t type,
653 boolean_t readonly, void *tag, objset_t **osp)
659 err = dsl_pool_hold(name, FTAG, &dp);
662 err = dsl_dataset_own(dp, name, tag, &ds);
664 dsl_pool_rele(dp, FTAG);
667 err = dmu_objset_own_impl(ds, type, readonly, tag, osp);
668 dsl_pool_rele(dp, FTAG);
670 if (err == 0 && dmu_objset_userobjspace_upgradable(*osp))
671 dmu_objset_userobjspace_upgrade(*osp);
677 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type,
678 boolean_t readonly, void *tag, objset_t **osp)
683 err = dsl_dataset_own_obj(dp, obj, tag, &ds);
687 return (dmu_objset_own_impl(ds, type, readonly, tag, osp));
691 dmu_objset_rele(objset_t *os, void *tag)
693 dsl_pool_t *dp = dmu_objset_pool(os);
694 dsl_dataset_rele(os->os_dsl_dataset, tag);
695 dsl_pool_rele(dp, tag);
699 * When we are called, os MUST refer to an objset associated with a dataset
700 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
701 * == tag. We will then release and reacquire ownership of the dataset while
702 * holding the pool config_rwlock to avoid intervening namespace or ownership
705 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
706 * release the hold on its dataset and acquire a new one on the dataset of the
707 * same name so that it can be partially torn down and reconstructed.
710 dmu_objset_refresh_ownership(objset_t *os, void *tag)
713 dsl_dataset_t *ds, *newds;
714 char name[ZFS_MAX_DATASET_NAME_LEN];
716 ds = os->os_dsl_dataset;
717 VERIFY3P(ds, !=, NULL);
718 VERIFY3P(ds->ds_owner, ==, tag);
719 VERIFY(dsl_dataset_long_held(ds));
721 dsl_dataset_name(ds, name);
722 dp = dmu_objset_pool(os);
723 dsl_pool_config_enter(dp, FTAG);
724 dmu_objset_disown(os, tag);
725 VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
726 VERIFY3P(newds, ==, os->os_dsl_dataset);
727 dsl_pool_config_exit(dp, FTAG);
731 dmu_objset_disown(objset_t *os, void *tag)
734 * Stop upgrading thread
736 dmu_objset_upgrade_stop(os);
737 dsl_dataset_disown(os->os_dsl_dataset, tag);
741 dmu_objset_evict_dbufs(objset_t *os)
746 dn_marker = kmem_alloc(sizeof (dnode_t), KM_SLEEP);
748 mutex_enter(&os->os_lock);
749 dn = list_head(&os->os_dnodes);
752 * Skip dnodes without holds. We have to do this dance
753 * because dnode_add_ref() only works if there is already a
754 * hold. If the dnode has no holds, then it has no dbufs.
756 if (dnode_add_ref(dn, FTAG)) {
757 list_insert_after(&os->os_dnodes, dn, dn_marker);
758 mutex_exit(&os->os_lock);
760 dnode_evict_dbufs(dn);
761 dnode_rele(dn, FTAG);
763 mutex_enter(&os->os_lock);
764 dn = list_next(&os->os_dnodes, dn_marker);
765 list_remove(&os->os_dnodes, dn_marker);
767 dn = list_next(&os->os_dnodes, dn);
770 mutex_exit(&os->os_lock);
772 kmem_free(dn_marker, sizeof (dnode_t));
774 if (DMU_USERUSED_DNODE(os) != NULL) {
775 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
776 dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
778 dnode_evict_dbufs(DMU_META_DNODE(os));
782 * Objset eviction processing is split into into two pieces.
783 * The first marks the objset as evicting, evicts any dbufs that
784 * have a refcount of zero, and then queues up the objset for the
785 * second phase of eviction. Once os->os_dnodes has been cleared by
786 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
787 * The second phase closes the special dnodes, dequeues the objset from
788 * the list of those undergoing eviction, and finally frees the objset.
790 * NOTE: Due to asynchronous eviction processing (invocation of
791 * dnode_buf_pageout()), it is possible for the meta dnode for the
792 * objset to have no holds even though os->os_dnodes is not empty.
795 dmu_objset_evict(objset_t *os)
799 dsl_dataset_t *ds = os->os_dsl_dataset;
801 for (t = 0; t < TXG_SIZE; t++)
802 ASSERT(!dmu_objset_is_dirty(os, t));
805 dsl_prop_unregister_all(ds, os);
810 dmu_objset_evict_dbufs(os);
812 mutex_enter(&os->os_lock);
813 spa_evicting_os_register(os->os_spa, os);
814 if (list_is_empty(&os->os_dnodes)) {
815 mutex_exit(&os->os_lock);
816 dmu_objset_evict_done(os);
818 mutex_exit(&os->os_lock);
823 dmu_objset_evict_done(objset_t *os)
825 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
827 dnode_special_close(&os->os_meta_dnode);
828 if (DMU_USERUSED_DNODE(os)) {
829 dnode_special_close(&os->os_userused_dnode);
830 dnode_special_close(&os->os_groupused_dnode);
832 zil_free(os->os_zil);
834 arc_buf_destroy(os->os_phys_buf, &os->os_phys_buf);
837 * This is a barrier to prevent the objset from going away in
838 * dnode_move() until we can safely ensure that the objset is still in
839 * use. We consider the objset valid before the barrier and invalid
842 rw_enter(&os_lock, RW_READER);
845 mutex_destroy(&os->os_lock);
846 mutex_destroy(&os->os_userused_lock);
847 mutex_destroy(&os->os_obj_lock);
848 mutex_destroy(&os->os_user_ptr_lock);
849 for (int i = 0; i < TXG_SIZE; i++) {
850 multilist_destroy(os->os_dirty_dnodes[i]);
852 spa_evicting_os_deregister(os->os_spa, os);
853 kmem_free(os, sizeof (objset_t));
857 dmu_objset_snap_cmtime(objset_t *os)
859 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
862 /* called from dsl for meta-objset */
864 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
865 dmu_objset_type_t type, dmu_tx_t *tx)
870 ASSERT(dmu_tx_is_syncing(tx));
873 VERIFY0(dmu_objset_from_ds(ds, &os));
875 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
877 mdn = DMU_META_DNODE(os);
879 dnode_allocate(mdn, DMU_OT_DNODE, DNODE_BLOCK_SIZE, DN_MAX_INDBLKSHIFT,
880 DMU_OT_NONE, 0, DNODE_MIN_SLOTS, tx);
883 * We don't want to have to increase the meta-dnode's nlevels
884 * later, because then we could do it in quescing context while
885 * we are also accessing it in open context.
887 * This precaution is not necessary for the MOS (ds == NULL),
888 * because the MOS is only updated in syncing context.
889 * This is most fortunate: the MOS is the only objset that
890 * needs to be synced multiple times as spa_sync() iterates
891 * to convergence, so minimizing its dn_nlevels matters.
897 * Determine the number of levels necessary for the meta-dnode
898 * to contain DN_MAX_OBJECT dnodes. Note that in order to
899 * ensure that we do not overflow 64 bits, there has to be
900 * a nlevels that gives us a number of blocks > DN_MAX_OBJECT
901 * but < 2^64. Therefore,
902 * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT) (10) must be
903 * less than (64 - log2(DN_MAX_OBJECT)) (16).
905 while ((uint64_t)mdn->dn_nblkptr <<
906 (mdn->dn_datablkshift - DNODE_SHIFT +
907 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
911 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
912 mdn->dn_nlevels = levels;
915 ASSERT(type != DMU_OST_NONE);
916 ASSERT(type != DMU_OST_ANY);
917 ASSERT(type < DMU_OST_NUMTYPES);
918 os->os_phys->os_type = type;
919 if (dmu_objset_userused_enabled(os)) {
920 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
921 if (dmu_objset_userobjused_enabled(os)) {
922 ds->ds_feature_activation_needed[
923 SPA_FEATURE_USEROBJ_ACCOUNTING] = B_TRUE;
924 os->os_phys->os_flags |=
925 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE;
927 os->os_flags = os->os_phys->os_flags;
930 dsl_dataset_dirty(ds, tx);
935 typedef struct dmu_objset_create_arg {
936 const char *doca_name;
938 void (*doca_userfunc)(objset_t *os, void *arg,
939 cred_t *cr, dmu_tx_t *tx);
941 dmu_objset_type_t doca_type;
943 } dmu_objset_create_arg_t;
947 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
949 dmu_objset_create_arg_t *doca = arg;
950 dsl_pool_t *dp = dmu_tx_pool(tx);
955 if (strchr(doca->doca_name, '@') != NULL)
956 return (SET_ERROR(EINVAL));
958 if (strlen(doca->doca_name) >= ZFS_MAX_DATASET_NAME_LEN)
959 return (SET_ERROR(ENAMETOOLONG));
961 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
965 dsl_dir_rele(pdd, FTAG);
966 return (SET_ERROR(EEXIST));
968 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
970 dsl_dir_rele(pdd, FTAG);
976 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
978 dmu_objset_create_arg_t *doca = arg;
979 dsl_pool_t *dp = dmu_tx_pool(tx);
987 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
989 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
990 doca->doca_cred, tx);
992 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
993 rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
994 bp = dsl_dataset_get_blkptr(ds);
995 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
996 ds, bp, doca->doca_type, tx);
997 rrw_exit(&ds->ds_bp_rwlock, FTAG);
999 if (doca->doca_userfunc != NULL) {
1000 doca->doca_userfunc(os, doca->doca_userarg,
1001 doca->doca_cred, tx);
1004 spa_history_log_internal_ds(ds, "create", tx, "");
1005 zvol_create_minors(dp->dp_spa, doca->doca_name, B_TRUE);
1007 dsl_dataset_rele(ds, FTAG);
1008 dsl_dir_rele(pdd, FTAG);
1012 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
1013 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
1015 dmu_objset_create_arg_t doca;
1017 doca.doca_name = name;
1018 doca.doca_cred = CRED();
1019 doca.doca_flags = flags;
1020 doca.doca_userfunc = func;
1021 doca.doca_userarg = arg;
1022 doca.doca_type = type;
1024 return (dsl_sync_task(name,
1025 dmu_objset_create_check, dmu_objset_create_sync, &doca,
1026 5, ZFS_SPACE_CHECK_NORMAL));
1029 typedef struct dmu_objset_clone_arg {
1030 const char *doca_clone;
1031 const char *doca_origin;
1033 } dmu_objset_clone_arg_t;
1037 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
1039 dmu_objset_clone_arg_t *doca = arg;
1043 dsl_dataset_t *origin;
1044 dsl_pool_t *dp = dmu_tx_pool(tx);
1046 if (strchr(doca->doca_clone, '@') != NULL)
1047 return (SET_ERROR(EINVAL));
1049 if (strlen(doca->doca_clone) >= ZFS_MAX_DATASET_NAME_LEN)
1050 return (SET_ERROR(ENAMETOOLONG));
1052 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
1056 dsl_dir_rele(pdd, FTAG);
1057 return (SET_ERROR(EEXIST));
1060 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
1063 dsl_dir_rele(pdd, FTAG);
1064 return (SET_ERROR(EDQUOT));
1066 dsl_dir_rele(pdd, FTAG);
1068 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
1072 /* You can only clone snapshots, not the head datasets. */
1073 if (!origin->ds_is_snapshot) {
1074 dsl_dataset_rele(origin, FTAG);
1075 return (SET_ERROR(EINVAL));
1077 dsl_dataset_rele(origin, FTAG);
1083 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
1085 dmu_objset_clone_arg_t *doca = arg;
1086 dsl_pool_t *dp = dmu_tx_pool(tx);
1089 dsl_dataset_t *origin, *ds;
1091 char namebuf[ZFS_MAX_DATASET_NAME_LEN];
1093 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
1094 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
1096 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
1097 doca->doca_cred, tx);
1099 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
1100 dsl_dataset_name(origin, namebuf);
1101 spa_history_log_internal_ds(ds, "clone", tx,
1102 "origin=%s (%llu)", namebuf, origin->ds_object);
1103 zvol_create_minors(dp->dp_spa, doca->doca_clone, B_TRUE);
1104 dsl_dataset_rele(ds, FTAG);
1105 dsl_dataset_rele(origin, FTAG);
1106 dsl_dir_rele(pdd, FTAG);
1110 dmu_objset_clone(const char *clone, const char *origin)
1112 dmu_objset_clone_arg_t doca;
1114 doca.doca_clone = clone;
1115 doca.doca_origin = origin;
1116 doca.doca_cred = CRED();
1118 return (dsl_sync_task(clone,
1119 dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
1120 5, ZFS_SPACE_CHECK_NORMAL));
1124 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
1127 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
1128 nvlist_t *snaps = fnvlist_alloc();
1130 fnvlist_add_boolean(snaps, longsnap);
1132 err = dsl_dataset_snapshot(snaps, NULL, NULL);
1133 fnvlist_free(snaps);
1138 dmu_objset_upgrade_task_cb(void *data)
1140 objset_t *os = data;
1142 mutex_enter(&os->os_upgrade_lock);
1143 os->os_upgrade_status = EINTR;
1144 if (!os->os_upgrade_exit) {
1145 mutex_exit(&os->os_upgrade_lock);
1147 os->os_upgrade_status = os->os_upgrade_cb(os);
1148 mutex_enter(&os->os_upgrade_lock);
1150 os->os_upgrade_exit = B_TRUE;
1151 os->os_upgrade_id = 0;
1152 mutex_exit(&os->os_upgrade_lock);
1156 dmu_objset_upgrade(objset_t *os, dmu_objset_upgrade_cb_t cb)
1158 if (os->os_upgrade_id != 0)
1161 mutex_enter(&os->os_upgrade_lock);
1162 if (os->os_upgrade_id == 0 && os->os_upgrade_status == 0) {
1163 os->os_upgrade_exit = B_FALSE;
1164 os->os_upgrade_cb = cb;
1165 os->os_upgrade_id = taskq_dispatch(
1166 os->os_spa->spa_upgrade_taskq,
1167 dmu_objset_upgrade_task_cb, os, TQ_SLEEP);
1168 if (os->os_upgrade_id == TASKQID_INVALID)
1169 os->os_upgrade_status = ENOMEM;
1171 mutex_exit(&os->os_upgrade_lock);
1175 dmu_objset_upgrade_stop(objset_t *os)
1177 mutex_enter(&os->os_upgrade_lock);
1178 os->os_upgrade_exit = B_TRUE;
1179 if (os->os_upgrade_id != 0) {
1180 taskqid_t id = os->os_upgrade_id;
1182 os->os_upgrade_id = 0;
1183 mutex_exit(&os->os_upgrade_lock);
1185 taskq_cancel_id(os->os_spa->spa_upgrade_taskq, id);
1187 mutex_exit(&os->os_upgrade_lock);
1192 dmu_objset_sync_dnodes(multilist_sublist_t *list, dmu_tx_t *tx)
1196 while ((dn = multilist_sublist_head(list)) != NULL) {
1197 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
1198 ASSERT(dn->dn_dbuf->db_data_pending);
1200 * Initialize dn_zio outside dnode_sync() because the
1201 * meta-dnode needs to set it ouside dnode_sync().
1203 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1206 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1207 multilist_sublist_remove(list, dn);
1209 multilist_t *newlist = dn->dn_objset->os_synced_dnodes;
1210 if (newlist != NULL) {
1211 (void) dnode_add_ref(dn, newlist);
1212 multilist_insert(newlist, dn);
1221 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1225 blkptr_t *bp = zio->io_bp;
1227 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1229 ASSERT(!BP_IS_EMBEDDED(bp));
1230 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1231 ASSERT0(BP_GET_LEVEL(bp));
1234 * Update rootbp fill count: it should be the number of objects
1235 * allocated in the object set (not counting the "special"
1236 * objects that are stored in the objset_phys_t -- the meta
1237 * dnode and user/group accounting objects).
1240 for (i = 0; i < dnp->dn_nblkptr; i++)
1241 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1242 if (os->os_dsl_dataset != NULL)
1243 rrw_enter(&os->os_dsl_dataset->ds_bp_rwlock, RW_WRITER, FTAG);
1244 *os->os_rootbp = *bp;
1245 if (os->os_dsl_dataset != NULL)
1246 rrw_exit(&os->os_dsl_dataset->ds_bp_rwlock, FTAG);
1251 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1253 blkptr_t *bp = zio->io_bp;
1254 blkptr_t *bp_orig = &zio->io_bp_orig;
1257 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1258 ASSERT(BP_EQUAL(bp, bp_orig));
1260 dsl_dataset_t *ds = os->os_dsl_dataset;
1261 dmu_tx_t *tx = os->os_synctx;
1263 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1264 dsl_dataset_block_born(ds, bp, tx);
1266 kmem_free(bp, sizeof (*bp));
1269 typedef struct sync_dnodes_arg {
1270 multilist_t *sda_list;
1271 int sda_sublist_idx;
1272 multilist_t *sda_newlist;
1274 } sync_dnodes_arg_t;
1277 sync_dnodes_task(void *arg)
1279 sync_dnodes_arg_t *sda = arg;
1281 multilist_sublist_t *ms =
1282 multilist_sublist_lock(sda->sda_list, sda->sda_sublist_idx);
1284 dmu_objset_sync_dnodes(ms, sda->sda_tx);
1286 multilist_sublist_unlock(ms);
1288 kmem_free(sda, sizeof (*sda));
1292 /* called from dsl */
1294 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1297 zbookmark_phys_t zb;
1301 dbuf_dirty_record_t *dr;
1302 blkptr_t *blkptr_copy = kmem_alloc(sizeof (*os->os_rootbp), KM_SLEEP);
1303 *blkptr_copy = *os->os_rootbp;
1305 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1307 ASSERT(dmu_tx_is_syncing(tx));
1308 /* XXX the write_done callback should really give us the tx... */
1311 if (os->os_dsl_dataset == NULL) {
1313 * This is the MOS. If we have upgraded,
1314 * spa_max_replication() could change, so reset
1317 os->os_copies = spa_max_replication(os->os_spa);
1321 * Create the root block IO
1323 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1324 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1325 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1326 arc_release(os->os_phys_buf, &os->os_phys_buf);
1328 dmu_write_policy(os, NULL, 0, 0, ZIO_COMPRESS_INHERIT, &zp);
1330 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1331 blkptr_copy, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1332 &zp, dmu_objset_write_ready, NULL, NULL, dmu_objset_write_done,
1333 os, ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
1336 * Sync special dnodes - the parent IO for the sync is the root block
1338 DMU_META_DNODE(os)->dn_zio = zio;
1339 dnode_sync(DMU_META_DNODE(os), tx);
1341 os->os_phys->os_flags = os->os_flags;
1343 if (DMU_USERUSED_DNODE(os) &&
1344 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1345 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1346 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1347 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1348 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1351 txgoff = tx->tx_txg & TXG_MASK;
1353 if (dmu_objset_userused_enabled(os)) {
1355 * We must create the list here because it uses the
1356 * dn_dirty_link[] of this txg. But it may already
1357 * exist because we call dsl_dataset_sync() twice per txg.
1359 if (os->os_synced_dnodes == NULL) {
1360 os->os_synced_dnodes =
1361 multilist_create(sizeof (dnode_t),
1362 offsetof(dnode_t, dn_dirty_link[txgoff]),
1363 dnode_multilist_index_func);
1365 ASSERT3U(os->os_synced_dnodes->ml_offset, ==,
1366 offsetof(dnode_t, dn_dirty_link[txgoff]));
1371 i < multilist_get_num_sublists(os->os_dirty_dnodes[txgoff]); i++) {
1372 sync_dnodes_arg_t *sda = kmem_alloc(sizeof (*sda), KM_SLEEP);
1373 sda->sda_list = os->os_dirty_dnodes[txgoff];
1374 sda->sda_sublist_idx = i;
1376 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
1377 sync_dnodes_task, sda, 0);
1378 /* callback frees sda */
1380 taskq_wait(dmu_objset_pool(os)->dp_sync_taskq);
1382 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1383 while ((dr = list_head(list)) != NULL) {
1384 ASSERT0(dr->dr_dbuf->db_level);
1385 list_remove(list, dr);
1387 zio_nowait(dr->dr_zio);
1390 /* Enable dnode backfill if enough objects have been freed. */
1391 if (os->os_freed_dnodes >= dmu_rescan_dnode_threshold) {
1392 os->os_rescan_dnodes = B_TRUE;
1393 os->os_freed_dnodes = 0;
1397 * Free intent log blocks up to this tx.
1399 zil_sync(os->os_zil, tx);
1400 os->os_phys->os_zil_header = os->os_zil_header;
1405 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1407 return (!multilist_is_empty(os->os_dirty_dnodes[txg & TXG_MASK]));
1410 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1413 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1419 dmu_objset_userused_enabled(objset_t *os)
1421 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1422 used_cbs[os->os_phys->os_type] != NULL &&
1423 DMU_USERUSED_DNODE(os) != NULL);
1427 dmu_objset_userobjused_enabled(objset_t *os)
1429 return (dmu_objset_userused_enabled(os) &&
1430 spa_feature_is_enabled(os->os_spa, SPA_FEATURE_USEROBJ_ACCOUNTING));
1433 typedef struct userquota_node {
1434 /* must be in the first filed, see userquota_update_cache() */
1435 char uqn_id[20 + DMU_OBJACCT_PREFIX_LEN];
1437 avl_node_t uqn_node;
1440 typedef struct userquota_cache {
1441 avl_tree_t uqc_user_deltas;
1442 avl_tree_t uqc_group_deltas;
1443 } userquota_cache_t;
1446 userquota_compare(const void *l, const void *r)
1448 const userquota_node_t *luqn = l;
1449 const userquota_node_t *ruqn = r;
1453 * NB: can only access uqn_id because userquota_update_cache() doesn't
1454 * pass in an entire userquota_node_t.
1456 rv = strcmp(luqn->uqn_id, ruqn->uqn_id);
1458 return (AVL_ISIGN(rv));
1462 do_userquota_cacheflush(objset_t *os, userquota_cache_t *cache, dmu_tx_t *tx)
1465 userquota_node_t *uqn;
1467 ASSERT(dmu_tx_is_syncing(tx));
1470 while ((uqn = avl_destroy_nodes(&cache->uqc_user_deltas,
1471 &cookie)) != NULL) {
1473 * os_userused_lock protects against concurrent calls to
1474 * zap_increment_int(). It's needed because zap_increment_int()
1475 * is not thread-safe (i.e. not atomic).
1477 mutex_enter(&os->os_userused_lock);
1478 VERIFY0(zap_increment(os, DMU_USERUSED_OBJECT,
1479 uqn->uqn_id, uqn->uqn_delta, tx));
1480 mutex_exit(&os->os_userused_lock);
1481 kmem_free(uqn, sizeof (*uqn));
1483 avl_destroy(&cache->uqc_user_deltas);
1486 while ((uqn = avl_destroy_nodes(&cache->uqc_group_deltas,
1487 &cookie)) != NULL) {
1488 mutex_enter(&os->os_userused_lock);
1489 VERIFY0(zap_increment(os, DMU_GROUPUSED_OBJECT,
1490 uqn->uqn_id, uqn->uqn_delta, tx));
1491 mutex_exit(&os->os_userused_lock);
1492 kmem_free(uqn, sizeof (*uqn));
1494 avl_destroy(&cache->uqc_group_deltas);
1498 userquota_update_cache(avl_tree_t *avl, const char *id, int64_t delta)
1500 userquota_node_t *uqn;
1503 ASSERT(strlen(id) < sizeof (uqn->uqn_id));
1505 * Use id directly for searching because uqn_id is the first field of
1506 * userquota_node_t and fields after uqn_id won't be accessed in
1509 uqn = avl_find(avl, (const void *)id, &idx);
1511 uqn = kmem_zalloc(sizeof (*uqn), KM_SLEEP);
1512 strlcpy(uqn->uqn_id, id, sizeof (uqn->uqn_id));
1513 avl_insert(avl, uqn, idx);
1515 uqn->uqn_delta += delta;
1519 do_userquota_update(userquota_cache_t *cache, uint64_t used, uint64_t flags,
1520 uint64_t user, uint64_t group, boolean_t subtract)
1522 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1523 int64_t delta = DNODE_MIN_SIZE + used;
1529 (void) sprintf(name, "%llx", (longlong_t)user);
1530 userquota_update_cache(&cache->uqc_user_deltas, name, delta);
1532 (void) sprintf(name, "%llx", (longlong_t)group);
1533 userquota_update_cache(&cache->uqc_group_deltas, name, delta);
1538 do_userobjquota_update(userquota_cache_t *cache, uint64_t flags,
1539 uint64_t user, uint64_t group, boolean_t subtract)
1541 if (flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) {
1542 char name[20 + DMU_OBJACCT_PREFIX_LEN];
1543 int delta = subtract ? -1 : 1;
1545 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx",
1547 userquota_update_cache(&cache->uqc_user_deltas, name, delta);
1549 (void) snprintf(name, sizeof (name), DMU_OBJACCT_PREFIX "%llx",
1551 userquota_update_cache(&cache->uqc_group_deltas, name, delta);
1555 typedef struct userquota_updates_arg {
1557 int uua_sublist_idx;
1559 } userquota_updates_arg_t;
1562 userquota_updates_task(void *arg)
1564 userquota_updates_arg_t *uua = arg;
1565 objset_t *os = uua->uua_os;
1566 dmu_tx_t *tx = uua->uua_tx;
1568 userquota_cache_t cache = { { 0 } };
1570 multilist_sublist_t *list =
1571 multilist_sublist_lock(os->os_synced_dnodes, uua->uua_sublist_idx);
1573 ASSERT(multilist_sublist_head(list) == NULL ||
1574 dmu_objset_userused_enabled(os));
1575 avl_create(&cache.uqc_user_deltas, userquota_compare,
1576 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1577 avl_create(&cache.uqc_group_deltas, userquota_compare,
1578 sizeof (userquota_node_t), offsetof(userquota_node_t, uqn_node));
1580 while ((dn = multilist_sublist_head(list)) != NULL) {
1582 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1583 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1584 dn->dn_phys->dn_flags &
1585 DNODE_FLAG_USERUSED_ACCOUNTED);
1587 flags = dn->dn_id_flags;
1589 if (flags & DN_ID_OLD_EXIST) {
1590 do_userquota_update(&cache,
1591 dn->dn_oldused, dn->dn_oldflags,
1592 dn->dn_olduid, dn->dn_oldgid, B_TRUE);
1593 do_userobjquota_update(&cache, dn->dn_oldflags,
1594 dn->dn_olduid, dn->dn_oldgid, B_TRUE);
1596 if (flags & DN_ID_NEW_EXIST) {
1597 do_userquota_update(&cache,
1598 DN_USED_BYTES(dn->dn_phys), dn->dn_phys->dn_flags,
1599 dn->dn_newuid, dn->dn_newgid, B_FALSE);
1600 do_userobjquota_update(&cache, dn->dn_phys->dn_flags,
1601 dn->dn_newuid, dn->dn_newgid, B_FALSE);
1604 mutex_enter(&dn->dn_mtx);
1606 dn->dn_oldflags = 0;
1607 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1608 dn->dn_olduid = dn->dn_newuid;
1609 dn->dn_oldgid = dn->dn_newgid;
1610 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1611 if (dn->dn_bonuslen == 0)
1612 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1614 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1616 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1617 mutex_exit(&dn->dn_mtx);
1619 multilist_sublist_remove(list, dn);
1620 dnode_rele(dn, os->os_synced_dnodes);
1622 do_userquota_cacheflush(os, &cache, tx);
1623 multilist_sublist_unlock(list);
1624 kmem_free(uua, sizeof (*uua));
1628 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1630 if (!dmu_objset_userused_enabled(os))
1633 /* Allocate the user/groupused objects if necessary. */
1634 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1635 VERIFY0(zap_create_claim(os,
1636 DMU_USERUSED_OBJECT,
1637 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1638 VERIFY0(zap_create_claim(os,
1639 DMU_GROUPUSED_OBJECT,
1640 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1644 i < multilist_get_num_sublists(os->os_synced_dnodes); i++) {
1645 userquota_updates_arg_t *uua =
1646 kmem_alloc(sizeof (*uua), KM_SLEEP);
1648 uua->uua_sublist_idx = i;
1650 /* note: caller does taskq_wait() */
1651 (void) taskq_dispatch(dmu_objset_pool(os)->dp_sync_taskq,
1652 userquota_updates_task, uua, 0);
1653 /* callback frees uua */
1658 * Returns a pointer to data to find uid/gid from
1660 * If a dirty record for transaction group that is syncing can't
1661 * be found then NULL is returned. In the NULL case it is assumed
1662 * the uid/gid aren't changing.
1665 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1667 dbuf_dirty_record_t *dr, **drp;
1670 if (db->db_dirtycnt == 0)
1671 return (db->db.db_data); /* Nothing is changing */
1673 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1674 if (dr->dr_txg == tx->tx_txg)
1682 DB_DNODE_ENTER(dr->dr_dbuf);
1683 dn = DB_DNODE(dr->dr_dbuf);
1685 if (dn->dn_bonuslen == 0 &&
1686 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1687 data = dr->dt.dl.dr_data->b_data;
1689 data = dr->dt.dl.dr_data;
1691 DB_DNODE_EXIT(dr->dr_dbuf);
1698 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1700 objset_t *os = dn->dn_objset;
1702 dmu_buf_impl_t *db = NULL;
1703 uint64_t *user = NULL;
1704 uint64_t *group = NULL;
1705 int flags = dn->dn_id_flags;
1707 boolean_t have_spill = B_FALSE;
1709 if (!dmu_objset_userused_enabled(dn->dn_objset))
1712 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1713 DN_ID_CHKED_SPILL)))
1716 if (before && dn->dn_bonuslen != 0)
1717 data = DN_BONUS(dn->dn_phys);
1718 else if (!before && dn->dn_bonuslen != 0) {
1721 mutex_enter(&db->db_mtx);
1722 data = dmu_objset_userquota_find_data(db, tx);
1724 data = DN_BONUS(dn->dn_phys);
1726 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1729 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1730 rf |= DB_RF_HAVESTRUCT;
1731 error = dmu_spill_hold_by_dnode(dn,
1732 rf | DB_RF_MUST_SUCCEED,
1733 FTAG, (dmu_buf_t **)&db);
1735 mutex_enter(&db->db_mtx);
1736 data = (before) ? db->db.db_data :
1737 dmu_objset_userquota_find_data(db, tx);
1738 have_spill = B_TRUE;
1740 mutex_enter(&dn->dn_mtx);
1741 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1742 mutex_exit(&dn->dn_mtx);
1748 user = &dn->dn_olduid;
1749 group = &dn->dn_oldgid;
1751 user = &dn->dn_newuid;
1752 group = &dn->dn_newgid;
1756 * Must always call the callback in case the object
1757 * type has changed and that type isn't an object type to track
1759 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1763 * Preserve existing uid/gid when the callback can't determine
1764 * what the new uid/gid are and the callback returned EEXIST.
1765 * The EEXIST error tells us to just use the existing uid/gid.
1766 * If we don't know what the old values are then just assign
1767 * them to 0, since that is a new file being created.
1769 if (!before && data == NULL && error == EEXIST) {
1770 if (flags & DN_ID_OLD_EXIST) {
1771 dn->dn_newuid = dn->dn_olduid;
1772 dn->dn_newgid = dn->dn_oldgid;
1781 mutex_exit(&db->db_mtx);
1783 mutex_enter(&dn->dn_mtx);
1784 if (error == 0 && before)
1785 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1786 if (error == 0 && !before)
1787 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1790 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1792 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1794 mutex_exit(&dn->dn_mtx);
1796 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1800 dmu_objset_userspace_present(objset_t *os)
1802 return (os->os_phys->os_flags &
1803 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1807 dmu_objset_userobjspace_present(objset_t *os)
1809 return (os->os_phys->os_flags &
1810 OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE);
1814 dmu_objset_space_upgrade(objset_t *os)
1820 * We simply need to mark every object dirty, so that it will be
1821 * synced out and now accounted. If this is called
1822 * concurrently, or if we already did some work before crashing,
1823 * that's fine, since we track each object's accounted state
1827 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1832 mutex_enter(&os->os_upgrade_lock);
1833 if (os->os_upgrade_exit)
1834 err = SET_ERROR(EINTR);
1835 mutex_exit(&os->os_upgrade_lock);
1839 if (issig(JUSTLOOKING) && issig(FORREAL))
1840 return (SET_ERROR(EINTR));
1842 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1845 tx = dmu_tx_create(os);
1846 dmu_tx_hold_bonus(tx, obj);
1847 objerr = dmu_tx_assign(tx, TXG_WAIT);
1852 dmu_buf_will_dirty(db, tx);
1853 dmu_buf_rele(db, FTAG);
1860 dmu_objset_userspace_upgrade(objset_t *os)
1864 if (dmu_objset_userspace_present(os))
1866 if (dmu_objset_is_snapshot(os))
1867 return (SET_ERROR(EINVAL));
1868 if (!dmu_objset_userused_enabled(os))
1869 return (SET_ERROR(ENOTSUP));
1871 err = dmu_objset_space_upgrade(os);
1875 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1876 txg_wait_synced(dmu_objset_pool(os), 0);
1881 dmu_objset_userobjspace_upgrade_cb(objset_t *os)
1885 if (dmu_objset_userobjspace_present(os))
1887 if (dmu_objset_is_snapshot(os))
1888 return (SET_ERROR(EINVAL));
1889 if (!dmu_objset_userobjused_enabled(os))
1890 return (SET_ERROR(ENOTSUP));
1892 dmu_objset_ds(os)->ds_feature_activation_needed[
1893 SPA_FEATURE_USEROBJ_ACCOUNTING] = B_TRUE;
1895 err = dmu_objset_space_upgrade(os);
1899 os->os_flags |= OBJSET_FLAG_USEROBJACCOUNTING_COMPLETE;
1900 txg_wait_synced(dmu_objset_pool(os), 0);
1905 dmu_objset_userobjspace_upgrade(objset_t *os)
1907 dmu_objset_upgrade(os, dmu_objset_userobjspace_upgrade_cb);
1911 dmu_objset_userobjspace_upgradable(objset_t *os)
1913 return (dmu_objset_type(os) == DMU_OST_ZFS &&
1914 !dmu_objset_is_snapshot(os) &&
1915 dmu_objset_userobjused_enabled(os) &&
1916 !dmu_objset_userobjspace_present(os));
1920 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1921 uint64_t *usedobjsp, uint64_t *availobjsp)
1923 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1924 usedobjsp, availobjsp);
1928 dmu_objset_fsid_guid(objset_t *os)
1930 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1934 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1936 stat->dds_type = os->os_phys->os_type;
1937 if (os->os_dsl_dataset)
1938 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1942 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1944 ASSERT(os->os_dsl_dataset ||
1945 os->os_phys->os_type == DMU_OST_META);
1947 if (os->os_dsl_dataset != NULL)
1948 dsl_dataset_stats(os->os_dsl_dataset, nv);
1950 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1951 os->os_phys->os_type);
1952 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1953 dmu_objset_userspace_present(os));
1957 dmu_objset_is_snapshot(objset_t *os)
1959 if (os->os_dsl_dataset != NULL)
1960 return (os->os_dsl_dataset->ds_is_snapshot);
1966 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1967 boolean_t *conflict)
1969 dsl_dataset_t *ds = os->os_dsl_dataset;
1972 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1973 return (SET_ERROR(ENOENT));
1975 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1976 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
1977 MT_NORMALIZE, real, maxlen, conflict));
1981 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1982 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1984 dsl_dataset_t *ds = os->os_dsl_dataset;
1985 zap_cursor_t cursor;
1986 zap_attribute_t attr;
1988 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1990 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1991 return (SET_ERROR(ENOENT));
1993 zap_cursor_init_serialized(&cursor,
1994 ds->ds_dir->dd_pool->dp_meta_objset,
1995 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
1997 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1998 zap_cursor_fini(&cursor);
1999 return (SET_ERROR(ENOENT));
2002 if (strlen(attr.za_name) + 1 > namelen) {
2003 zap_cursor_fini(&cursor);
2004 return (SET_ERROR(ENAMETOOLONG));
2007 (void) strcpy(name, attr.za_name);
2009 *idp = attr.za_first_integer;
2011 *case_conflict = attr.za_normalization_conflict;
2012 zap_cursor_advance(&cursor);
2013 *offp = zap_cursor_serialize(&cursor);
2014 zap_cursor_fini(&cursor);
2020 dmu_snapshot_lookup(objset_t *os, const char *name, uint64_t *value)
2022 return (dsl_dataset_snap_lookup(os->os_dsl_dataset, name, value));
2026 dmu_dir_list_next(objset_t *os, int namelen, char *name,
2027 uint64_t *idp, uint64_t *offp)
2029 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
2030 zap_cursor_t cursor;
2031 zap_attribute_t attr;
2033 /* there is no next dir on a snapshot! */
2034 if (os->os_dsl_dataset->ds_object !=
2035 dsl_dir_phys(dd)->dd_head_dataset_obj)
2036 return (SET_ERROR(ENOENT));
2038 zap_cursor_init_serialized(&cursor,
2039 dd->dd_pool->dp_meta_objset,
2040 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
2042 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
2043 zap_cursor_fini(&cursor);
2044 return (SET_ERROR(ENOENT));
2047 if (strlen(attr.za_name) + 1 > namelen) {
2048 zap_cursor_fini(&cursor);
2049 return (SET_ERROR(ENAMETOOLONG));
2052 (void) strcpy(name, attr.za_name);
2054 *idp = attr.za_first_integer;
2055 zap_cursor_advance(&cursor);
2056 *offp = zap_cursor_serialize(&cursor);
2057 zap_cursor_fini(&cursor);
2062 typedef struct dmu_objset_find_ctx {
2066 char *dc_ddname; /* last component of ddobj's name */
2067 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *);
2070 kmutex_t *dc_error_lock;
2072 } dmu_objset_find_ctx_t;
2075 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp)
2077 dsl_pool_t *dp = dcp->dc_dp;
2078 dmu_objset_find_ctx_t *child_dcp;
2082 zap_attribute_t *attr;
2086 /* don't process if there already was an error */
2087 if (*dcp->dc_error != 0)
2091 * Note: passing the name (dc_ddname) here is optional, but it
2092 * improves performance because we don't need to call
2093 * zap_value_search() to determine the name.
2095 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, dcp->dc_ddname, FTAG, &dd);
2099 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2100 if (dd->dd_myname[0] == '$') {
2101 dsl_dir_rele(dd, FTAG);
2105 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
2106 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
2109 * Iterate over all children.
2111 if (dcp->dc_flags & DS_FIND_CHILDREN) {
2112 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2113 dsl_dir_phys(dd)->dd_child_dir_zapobj);
2114 zap_cursor_retrieve(&zc, attr) == 0;
2115 (void) zap_cursor_advance(&zc)) {
2116 ASSERT3U(attr->za_integer_length, ==,
2118 ASSERT3U(attr->za_num_integers, ==, 1);
2121 kmem_alloc(sizeof (*child_dcp), KM_SLEEP);
2123 child_dcp->dc_ddobj = attr->za_first_integer;
2124 child_dcp->dc_ddname = spa_strdup(attr->za_name);
2125 if (dcp->dc_tq != NULL)
2126 (void) taskq_dispatch(dcp->dc_tq,
2127 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP);
2129 dmu_objset_find_dp_impl(child_dcp);
2131 zap_cursor_fini(&zc);
2135 * Iterate over all snapshots.
2137 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) {
2139 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2144 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2145 dsl_dataset_rele(ds, FTAG);
2147 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2148 zap_cursor_retrieve(&zc, attr) == 0;
2149 (void) zap_cursor_advance(&zc)) {
2150 ASSERT3U(attr->za_integer_length, ==,
2152 ASSERT3U(attr->za_num_integers, ==, 1);
2154 err = dsl_dataset_hold_obj(dp,
2155 attr->za_first_integer, FTAG, &ds);
2158 err = dcp->dc_func(dp, ds, dcp->dc_arg);
2159 dsl_dataset_rele(ds, FTAG);
2163 zap_cursor_fini(&zc);
2167 kmem_free(attr, sizeof (zap_attribute_t));
2170 dsl_dir_rele(dd, FTAG);
2177 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2180 * Note: we hold the dir while calling dsl_dataset_hold_obj() so
2181 * that the dir will remain cached, and we won't have to re-instantiate
2182 * it (which could be expensive due to finding its name via
2183 * zap_value_search()).
2185 dsl_dir_rele(dd, FTAG);
2188 err = dcp->dc_func(dp, ds, dcp->dc_arg);
2189 dsl_dataset_rele(ds, FTAG);
2193 mutex_enter(dcp->dc_error_lock);
2194 /* only keep first error */
2195 if (*dcp->dc_error == 0)
2196 *dcp->dc_error = err;
2197 mutex_exit(dcp->dc_error_lock);
2200 if (dcp->dc_ddname != NULL)
2201 spa_strfree(dcp->dc_ddname);
2202 kmem_free(dcp, sizeof (*dcp));
2206 dmu_objset_find_dp_cb(void *arg)
2208 dmu_objset_find_ctx_t *dcp = arg;
2209 dsl_pool_t *dp = dcp->dc_dp;
2212 * We need to get a pool_config_lock here, as there are several
2213 * asssert(pool_config_held) down the stack. Getting a lock via
2214 * dsl_pool_config_enter is risky, as it might be stalled by a
2215 * pending writer. This would deadlock, as the write lock can
2216 * only be granted when our parent thread gives up the lock.
2217 * The _prio interface gives us priority over a pending writer.
2219 dsl_pool_config_enter_prio(dp, FTAG);
2221 dmu_objset_find_dp_impl(dcp);
2223 dsl_pool_config_exit(dp, FTAG);
2227 * Find objsets under and including ddobj, call func(ds) on each.
2228 * The order for the enumeration is completely undefined.
2229 * func is called with dsl_pool_config held.
2232 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
2233 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
2238 dmu_objset_find_ctx_t *dcp;
2241 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL);
2242 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP);
2245 dcp->dc_ddobj = ddobj;
2246 dcp->dc_ddname = NULL;
2247 dcp->dc_func = func;
2249 dcp->dc_flags = flags;
2250 dcp->dc_error_lock = &err_lock;
2251 dcp->dc_error = &error;
2253 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) {
2255 * In case a write lock is held we can't make use of
2256 * parallelism, as down the stack of the worker threads
2257 * the lock is asserted via dsl_pool_config_held.
2258 * In case of a read lock this is solved by getting a read
2259 * lock in each worker thread, which isn't possible in case
2260 * of a writer lock. So we fall back to the synchronous path
2262 * In the future it might be possible to get some magic into
2263 * dsl_pool_config_held in a way that it returns true for
2264 * the worker threads so that a single lock held from this
2265 * thread suffices. For now, stay single threaded.
2267 dmu_objset_find_dp_impl(dcp);
2268 mutex_destroy(&err_lock);
2273 ntasks = dmu_find_threads;
2275 ntasks = vdev_count_leaves(dp->dp_spa) * 4;
2276 tq = taskq_create("dmu_objset_find", ntasks, maxclsyspri, ntasks,
2279 kmem_free(dcp, sizeof (*dcp));
2280 mutex_destroy(&err_lock);
2282 return (SET_ERROR(ENOMEM));
2286 /* dcp will be freed by task */
2287 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP);
2290 * PORTING: this code relies on the property of taskq_wait to wait
2291 * until no more tasks are queued and no more tasks are active. As
2292 * we always queue new tasks from within other tasks, task_wait
2293 * reliably waits for the full recursion to finish, even though we
2294 * enqueue new tasks after taskq_wait has been called.
2295 * On platforms other than illumos, taskq_wait may not have this
2300 mutex_destroy(&err_lock);
2306 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
2307 * The dp_config_rwlock must not be held when this is called, and it
2308 * will not be held when the callback is called.
2309 * Therefore this function should only be used when the pool is not changing
2310 * (e.g. in syncing context), or the callback can deal with the possible races.
2313 dmu_objset_find_impl(spa_t *spa, const char *name,
2314 int func(const char *, void *), void *arg, int flags)
2317 dsl_pool_t *dp = spa_get_dsl(spa);
2320 zap_attribute_t *attr;
2325 dsl_pool_config_enter(dp, FTAG);
2327 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
2329 dsl_pool_config_exit(dp, FTAG);
2333 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
2334 if (dd->dd_myname[0] == '$') {
2335 dsl_dir_rele(dd, FTAG);
2336 dsl_pool_config_exit(dp, FTAG);
2340 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
2341 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
2344 * Iterate over all children.
2346 if (flags & DS_FIND_CHILDREN) {
2347 for (zap_cursor_init(&zc, dp->dp_meta_objset,
2348 dsl_dir_phys(dd)->dd_child_dir_zapobj);
2349 zap_cursor_retrieve(&zc, attr) == 0;
2350 (void) zap_cursor_advance(&zc)) {
2351 ASSERT3U(attr->za_integer_length, ==,
2353 ASSERT3U(attr->za_num_integers, ==, 1);
2355 child = kmem_asprintf("%s/%s", name, attr->za_name);
2356 dsl_pool_config_exit(dp, FTAG);
2357 err = dmu_objset_find_impl(spa, child,
2359 dsl_pool_config_enter(dp, FTAG);
2364 zap_cursor_fini(&zc);
2367 dsl_dir_rele(dd, FTAG);
2368 dsl_pool_config_exit(dp, FTAG);
2369 kmem_free(attr, sizeof (zap_attribute_t));
2375 * Iterate over all snapshots.
2377 if (flags & DS_FIND_SNAPSHOTS) {
2378 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
2383 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
2384 dsl_dataset_rele(ds, FTAG);
2386 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
2387 zap_cursor_retrieve(&zc, attr) == 0;
2388 (void) zap_cursor_advance(&zc)) {
2389 ASSERT3U(attr->za_integer_length, ==,
2391 ASSERT3U(attr->za_num_integers, ==, 1);
2393 child = kmem_asprintf("%s@%s",
2394 name, attr->za_name);
2395 dsl_pool_config_exit(dp, FTAG);
2396 err = func(child, arg);
2397 dsl_pool_config_enter(dp, FTAG);
2402 zap_cursor_fini(&zc);
2406 dsl_dir_rele(dd, FTAG);
2407 kmem_free(attr, sizeof (zap_attribute_t));
2408 dsl_pool_config_exit(dp, FTAG);
2413 /* Apply to self. */
2414 return (func(name, arg));
2418 * See comment above dmu_objset_find_impl().
2421 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
2427 error = spa_open(name, &spa, FTAG);
2430 error = dmu_objset_find_impl(spa, name, func, arg, flags);
2431 spa_close(spa, FTAG);
2436 dmu_objset_set_user(objset_t *os, void *user_ptr)
2438 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2439 os->os_user_ptr = user_ptr;
2443 dmu_objset_get_user(objset_t *os)
2445 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
2446 return (os->os_user_ptr);
2450 * Determine name of filesystem, given name of snapshot.
2451 * buf must be at least ZFS_MAX_DATASET_NAME_LEN bytes
2454 dmu_fsname(const char *snapname, char *buf)
2456 char *atp = strchr(snapname, '@');
2458 return (SET_ERROR(EINVAL));
2459 if (atp - snapname >= ZFS_MAX_DATASET_NAME_LEN)
2460 return (SET_ERROR(ENAMETOOLONG));
2461 (void) strlcpy(buf, snapname, atp - snapname + 1);
2466 * Call when we think we're going to write/free space in open context to track
2467 * the amount of dirty data in the open txg, which is also the amount
2468 * of memory that can not be evicted until this txg syncs.
2471 dmu_objset_willuse_space(objset_t *os, int64_t space, dmu_tx_t *tx)
2473 dsl_dataset_t *ds = os->os_dsl_dataset;
2474 int64_t aspace = spa_get_worst_case_asize(os->os_spa, space);
2477 dsl_dir_willuse_space(ds->ds_dir, aspace, tx);
2478 dsl_pool_dirty_space(dmu_tx_pool(tx), space, tx);
2482 #if defined(_KERNEL) && defined(HAVE_SPL)
2483 EXPORT_SYMBOL(dmu_objset_zil);
2484 EXPORT_SYMBOL(dmu_objset_pool);
2485 EXPORT_SYMBOL(dmu_objset_ds);
2486 EXPORT_SYMBOL(dmu_objset_type);
2487 EXPORT_SYMBOL(dmu_objset_name);
2488 EXPORT_SYMBOL(dmu_objset_hold);
2489 EXPORT_SYMBOL(dmu_objset_own);
2490 EXPORT_SYMBOL(dmu_objset_rele);
2491 EXPORT_SYMBOL(dmu_objset_disown);
2492 EXPORT_SYMBOL(dmu_objset_from_ds);
2493 EXPORT_SYMBOL(dmu_objset_create);
2494 EXPORT_SYMBOL(dmu_objset_clone);
2495 EXPORT_SYMBOL(dmu_objset_stats);
2496 EXPORT_SYMBOL(dmu_objset_fast_stat);
2497 EXPORT_SYMBOL(dmu_objset_spa);
2498 EXPORT_SYMBOL(dmu_objset_space);
2499 EXPORT_SYMBOL(dmu_objset_fsid_guid);
2500 EXPORT_SYMBOL(dmu_objset_find);
2501 EXPORT_SYMBOL(dmu_objset_byteswap);
2502 EXPORT_SYMBOL(dmu_objset_evict_dbufs);
2503 EXPORT_SYMBOL(dmu_objset_snap_cmtime);
2504 EXPORT_SYMBOL(dmu_objset_dnodesize);
2506 EXPORT_SYMBOL(dmu_objset_sync);
2507 EXPORT_SYMBOL(dmu_objset_is_dirty);
2508 EXPORT_SYMBOL(dmu_objset_create_impl);
2509 EXPORT_SYMBOL(dmu_objset_open_impl);
2510 EXPORT_SYMBOL(dmu_objset_evict);
2511 EXPORT_SYMBOL(dmu_objset_register_type);
2512 EXPORT_SYMBOL(dmu_objset_do_userquota_updates);
2513 EXPORT_SYMBOL(dmu_objset_userquota_get_ids);
2514 EXPORT_SYMBOL(dmu_objset_userused_enabled);
2515 EXPORT_SYMBOL(dmu_objset_userspace_upgrade);
2516 EXPORT_SYMBOL(dmu_objset_userspace_present);
2517 EXPORT_SYMBOL(dmu_objset_userobjused_enabled);
2518 EXPORT_SYMBOL(dmu_objset_userobjspace_upgrade);
2519 EXPORT_SYMBOL(dmu_objset_userobjspace_upgradable);
2520 EXPORT_SYMBOL(dmu_objset_userobjspace_present);