4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2013 by Saso Kiselkov. All rights reserved.
25 * Copyright (c) 2013, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
27 * Copyright 2015 Nexenta Systems, Inc. All rights reserved.
28 * Copyright (c) 2015, STRATO AG, Inc. All rights reserved.
31 /* Portions Copyright 2010 Robert Milkowski */
34 #include <sys/zfs_context.h>
35 #include <sys/dmu_objset.h>
36 #include <sys/dsl_dir.h>
37 #include <sys/dsl_dataset.h>
38 #include <sys/dsl_prop.h>
39 #include <sys/dsl_pool.h>
40 #include <sys/dsl_synctask.h>
41 #include <sys/dsl_deleg.h>
42 #include <sys/dnode.h>
45 #include <sys/dmu_tx.h>
48 #include <sys/dmu_impl.h>
49 #include <sys/zfs_ioctl.h>
51 #include <sys/zfs_onexit.h>
52 #include <sys/dsl_destroy.h>
56 * Needed to close a window in dnode_move() that allows the objset to be freed
57 * before it can be safely accessed.
62 * Tunable to overwrite the maximum number of threads for the parallization
63 * of dmu_objset_find_dp, needed to speed up the import of pools with many
65 * Default is 4 times the number of leaf vdevs.
67 int dmu_find_threads = 0;
69 static void dmu_objset_find_dp_cb(void *arg);
74 rw_init(&os_lock, NULL, RW_DEFAULT, NULL);
84 dmu_objset_spa(objset_t *os)
90 dmu_objset_zil(objset_t *os)
96 dmu_objset_pool(objset_t *os)
100 if ((ds = os->os_dsl_dataset) != NULL && ds->ds_dir)
101 return (ds->ds_dir->dd_pool);
103 return (spa_get_dsl(os->os_spa));
107 dmu_objset_ds(objset_t *os)
109 return (os->os_dsl_dataset);
113 dmu_objset_type(objset_t *os)
115 return (os->os_phys->os_type);
119 dmu_objset_name(objset_t *os, char *buf)
121 dsl_dataset_name(os->os_dsl_dataset, buf);
125 dmu_objset_id(objset_t *os)
127 dsl_dataset_t *ds = os->os_dsl_dataset;
129 return (ds ? ds->ds_object : 0);
133 dmu_objset_syncprop(objset_t *os)
135 return (os->os_sync);
139 dmu_objset_logbias(objset_t *os)
141 return (os->os_logbias);
145 checksum_changed_cb(void *arg, uint64_t newval)
150 * Inheritance should have been done by now.
152 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
154 os->os_checksum = zio_checksum_select(newval, ZIO_CHECKSUM_ON_VALUE);
158 compression_changed_cb(void *arg, uint64_t newval)
163 * Inheritance and range checking should have been done by now.
165 ASSERT(newval != ZIO_COMPRESS_INHERIT);
167 os->os_compress = zio_compress_select(os->os_spa, newval,
172 copies_changed_cb(void *arg, uint64_t newval)
177 * Inheritance and range checking should have been done by now.
180 ASSERT(newval <= spa_max_replication(os->os_spa));
182 os->os_copies = newval;
186 dedup_changed_cb(void *arg, uint64_t newval)
189 spa_t *spa = os->os_spa;
190 enum zio_checksum checksum;
193 * Inheritance should have been done by now.
195 ASSERT(newval != ZIO_CHECKSUM_INHERIT);
197 checksum = zio_checksum_dedup_select(spa, newval, ZIO_CHECKSUM_OFF);
199 os->os_dedup_checksum = checksum & ZIO_CHECKSUM_MASK;
200 os->os_dedup_verify = !!(checksum & ZIO_CHECKSUM_VERIFY);
204 primary_cache_changed_cb(void *arg, uint64_t newval)
209 * Inheritance and range checking should have been done by now.
211 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
212 newval == ZFS_CACHE_METADATA);
214 os->os_primary_cache = newval;
218 secondary_cache_changed_cb(void *arg, uint64_t newval)
223 * Inheritance and range checking should have been done by now.
225 ASSERT(newval == ZFS_CACHE_ALL || newval == ZFS_CACHE_NONE ||
226 newval == ZFS_CACHE_METADATA);
228 os->os_secondary_cache = newval;
232 sync_changed_cb(void *arg, uint64_t newval)
237 * Inheritance and range checking should have been done by now.
239 ASSERT(newval == ZFS_SYNC_STANDARD || newval == ZFS_SYNC_ALWAYS ||
240 newval == ZFS_SYNC_DISABLED);
242 os->os_sync = newval;
244 zil_set_sync(os->os_zil, newval);
248 redundant_metadata_changed_cb(void *arg, uint64_t newval)
253 * Inheritance and range checking should have been done by now.
255 ASSERT(newval == ZFS_REDUNDANT_METADATA_ALL ||
256 newval == ZFS_REDUNDANT_METADATA_MOST);
258 os->os_redundant_metadata = newval;
262 logbias_changed_cb(void *arg, uint64_t newval)
266 ASSERT(newval == ZFS_LOGBIAS_LATENCY ||
267 newval == ZFS_LOGBIAS_THROUGHPUT);
268 os->os_logbias = newval;
270 zil_set_logbias(os->os_zil, newval);
274 recordsize_changed_cb(void *arg, uint64_t newval)
278 os->os_recordsize = newval;
282 dmu_objset_byteswap(void *buf, size_t size)
284 objset_phys_t *osp = buf;
286 ASSERT(size == OBJSET_OLD_PHYS_SIZE || size == sizeof (objset_phys_t));
287 dnode_byteswap(&osp->os_meta_dnode);
288 byteswap_uint64_array(&osp->os_zil_header, sizeof (zil_header_t));
289 osp->os_type = BSWAP_64(osp->os_type);
290 osp->os_flags = BSWAP_64(osp->os_flags);
291 if (size == sizeof (objset_phys_t)) {
292 dnode_byteswap(&osp->os_userused_dnode);
293 dnode_byteswap(&osp->os_groupused_dnode);
298 dmu_objset_open_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
304 ASSERT(ds == NULL || MUTEX_HELD(&ds->ds_opening_lock));
306 os = kmem_zalloc(sizeof (objset_t), KM_SLEEP);
307 os->os_dsl_dataset = ds;
310 if (!BP_IS_HOLE(os->os_rootbp)) {
311 arc_flags_t aflags = ARC_FLAG_WAIT;
313 SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
314 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
316 if (DMU_OS_IS_L2CACHEABLE(os))
317 aflags |= ARC_FLAG_L2CACHE;
318 if (DMU_OS_IS_L2COMPRESSIBLE(os))
319 aflags |= ARC_FLAG_L2COMPRESS;
321 dprintf_bp(os->os_rootbp, "reading %s", "");
322 err = arc_read(NULL, spa, os->os_rootbp,
323 arc_getbuf_func, &os->os_phys_buf,
324 ZIO_PRIORITY_SYNC_READ, ZIO_FLAG_CANFAIL, &aflags, &zb);
326 kmem_free(os, sizeof (objset_t));
327 /* convert checksum errors into IO errors */
329 err = SET_ERROR(EIO);
333 /* Increase the blocksize if we are permitted. */
334 if (spa_version(spa) >= SPA_VERSION_USERSPACE &&
335 arc_buf_size(os->os_phys_buf) < sizeof (objset_phys_t)) {
336 arc_buf_t *buf = arc_buf_alloc(spa,
337 sizeof (objset_phys_t), &os->os_phys_buf,
339 bzero(buf->b_data, sizeof (objset_phys_t));
340 bcopy(os->os_phys_buf->b_data, buf->b_data,
341 arc_buf_size(os->os_phys_buf));
342 (void) arc_buf_remove_ref(os->os_phys_buf,
344 os->os_phys_buf = buf;
347 os->os_phys = os->os_phys_buf->b_data;
348 os->os_flags = os->os_phys->os_flags;
350 int size = spa_version(spa) >= SPA_VERSION_USERSPACE ?
351 sizeof (objset_phys_t) : OBJSET_OLD_PHYS_SIZE;
352 os->os_phys_buf = arc_buf_alloc(spa, size,
353 &os->os_phys_buf, ARC_BUFC_METADATA);
354 os->os_phys = os->os_phys_buf->b_data;
355 bzero(os->os_phys, size);
359 * Note: the changed_cb will be called once before the register
360 * func returns, thus changing the checksum/compression from the
361 * default (fletcher2/off). Snapshots don't need to know about
362 * checksum/compression/copies.
365 err = dsl_prop_register(ds,
366 zfs_prop_to_name(ZFS_PROP_PRIMARYCACHE),
367 primary_cache_changed_cb, os);
369 err = dsl_prop_register(ds,
370 zfs_prop_to_name(ZFS_PROP_SECONDARYCACHE),
371 secondary_cache_changed_cb, os);
373 if (!ds->ds_is_snapshot) {
375 err = dsl_prop_register(ds,
376 zfs_prop_to_name(ZFS_PROP_CHECKSUM),
377 checksum_changed_cb, os);
380 err = dsl_prop_register(ds,
381 zfs_prop_to_name(ZFS_PROP_COMPRESSION),
382 compression_changed_cb, os);
385 err = dsl_prop_register(ds,
386 zfs_prop_to_name(ZFS_PROP_COPIES),
387 copies_changed_cb, os);
390 err = dsl_prop_register(ds,
391 zfs_prop_to_name(ZFS_PROP_DEDUP),
392 dedup_changed_cb, os);
395 err = dsl_prop_register(ds,
396 zfs_prop_to_name(ZFS_PROP_LOGBIAS),
397 logbias_changed_cb, os);
400 err = dsl_prop_register(ds,
401 zfs_prop_to_name(ZFS_PROP_SYNC),
402 sync_changed_cb, os);
405 err = dsl_prop_register(ds,
407 ZFS_PROP_REDUNDANT_METADATA),
408 redundant_metadata_changed_cb, os);
411 err = dsl_prop_register(ds,
412 zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
413 recordsize_changed_cb, os);
417 VERIFY(arc_buf_remove_ref(os->os_phys_buf,
419 kmem_free(os, sizeof (objset_t));
423 /* It's the meta-objset. */
424 os->os_checksum = ZIO_CHECKSUM_FLETCHER_4;
425 os->os_compress = ZIO_COMPRESS_ON;
426 os->os_copies = spa_max_replication(spa);
427 os->os_dedup_checksum = ZIO_CHECKSUM_OFF;
428 os->os_dedup_verify = B_FALSE;
429 os->os_logbias = ZFS_LOGBIAS_LATENCY;
430 os->os_sync = ZFS_SYNC_STANDARD;
431 os->os_primary_cache = ZFS_CACHE_ALL;
432 os->os_secondary_cache = ZFS_CACHE_ALL;
435 if (ds == NULL || !ds->ds_is_snapshot)
436 os->os_zil_header = os->os_phys->os_zil_header;
437 os->os_zil = zil_alloc(os, &os->os_zil_header);
439 for (i = 0; i < TXG_SIZE; i++) {
440 list_create(&os->os_dirty_dnodes[i], sizeof (dnode_t),
441 offsetof(dnode_t, dn_dirty_link[i]));
442 list_create(&os->os_free_dnodes[i], sizeof (dnode_t),
443 offsetof(dnode_t, dn_dirty_link[i]));
445 list_create(&os->os_dnodes, sizeof (dnode_t),
446 offsetof(dnode_t, dn_link));
447 list_create(&os->os_downgraded_dbufs, sizeof (dmu_buf_impl_t),
448 offsetof(dmu_buf_impl_t, db_link));
450 mutex_init(&os->os_lock, NULL, MUTEX_DEFAULT, NULL);
451 mutex_init(&os->os_obj_lock, NULL, MUTEX_DEFAULT, NULL);
452 mutex_init(&os->os_user_ptr_lock, NULL, MUTEX_DEFAULT, NULL);
454 dnode_special_open(os, &os->os_phys->os_meta_dnode,
455 DMU_META_DNODE_OBJECT, &os->os_meta_dnode);
456 if (arc_buf_size(os->os_phys_buf) >= sizeof (objset_phys_t)) {
457 dnode_special_open(os, &os->os_phys->os_userused_dnode,
458 DMU_USERUSED_OBJECT, &os->os_userused_dnode);
459 dnode_special_open(os, &os->os_phys->os_groupused_dnode,
460 DMU_GROUPUSED_OBJECT, &os->os_groupused_dnode);
468 dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
472 mutex_enter(&ds->ds_opening_lock);
473 if (ds->ds_objset == NULL) {
475 err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
476 ds, dsl_dataset_get_blkptr(ds), &os);
479 mutex_enter(&ds->ds_lock);
480 ASSERT(ds->ds_objset == NULL);
482 mutex_exit(&ds->ds_lock);
485 *osp = ds->ds_objset;
486 mutex_exit(&ds->ds_opening_lock);
491 * Holds the pool while the objset is held. Therefore only one objset
492 * can be held at a time.
495 dmu_objset_hold(const char *name, void *tag, objset_t **osp)
501 err = dsl_pool_hold(name, tag, &dp);
504 err = dsl_dataset_hold(dp, name, tag, &ds);
506 dsl_pool_rele(dp, tag);
510 err = dmu_objset_from_ds(ds, osp);
512 dsl_dataset_rele(ds, tag);
513 dsl_pool_rele(dp, tag);
520 dmu_objset_own_impl(dsl_dataset_t *ds, dmu_objset_type_t type,
521 boolean_t readonly, void *tag, objset_t **osp)
525 err = dmu_objset_from_ds(ds, osp);
527 dsl_dataset_disown(ds, tag);
528 } else if (type != DMU_OST_ANY && type != (*osp)->os_phys->os_type) {
529 dsl_dataset_disown(ds, tag);
530 return (SET_ERROR(EINVAL));
531 } else if (!readonly && dsl_dataset_is_snapshot(ds)) {
532 dsl_dataset_disown(ds, tag);
533 return (SET_ERROR(EROFS));
539 * dsl_pool must not be held when this is called.
540 * Upon successful return, there will be a longhold on the dataset,
541 * and the dsl_pool will not be held.
544 dmu_objset_own(const char *name, dmu_objset_type_t type,
545 boolean_t readonly, void *tag, objset_t **osp)
551 err = dsl_pool_hold(name, FTAG, &dp);
554 err = dsl_dataset_own(dp, name, tag, &ds);
556 dsl_pool_rele(dp, FTAG);
559 err = dmu_objset_own_impl(ds, type, readonly, tag, osp);
560 dsl_pool_rele(dp, FTAG);
566 dmu_objset_own_obj(dsl_pool_t *dp, uint64_t obj, dmu_objset_type_t type,
567 boolean_t readonly, void *tag, objset_t **osp)
572 err = dsl_dataset_own_obj(dp, obj, tag, &ds);
576 return (dmu_objset_own_impl(ds, type, readonly, tag, osp));
580 dmu_objset_rele(objset_t *os, void *tag)
582 dsl_pool_t *dp = dmu_objset_pool(os);
583 dsl_dataset_rele(os->os_dsl_dataset, tag);
584 dsl_pool_rele(dp, tag);
588 * When we are called, os MUST refer to an objset associated with a dataset
589 * that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
590 * == tag. We will then release and reacquire ownership of the dataset while
591 * holding the pool config_rwlock to avoid intervening namespace or ownership
594 * This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
595 * release the hold on its dataset and acquire a new one on the dataset of the
596 * same name so that it can be partially torn down and reconstructed.
599 dmu_objset_refresh_ownership(objset_t *os, void *tag)
602 dsl_dataset_t *ds, *newds;
603 char name[MAXNAMELEN];
605 ds = os->os_dsl_dataset;
606 VERIFY3P(ds, !=, NULL);
607 VERIFY3P(ds->ds_owner, ==, tag);
608 VERIFY(dsl_dataset_long_held(ds));
610 dsl_dataset_name(ds, name);
611 dp = dmu_objset_pool(os);
612 dsl_pool_config_enter(dp, FTAG);
613 dmu_objset_disown(os, tag);
614 VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
615 VERIFY3P(newds, ==, os->os_dsl_dataset);
616 dsl_pool_config_exit(dp, FTAG);
620 dmu_objset_disown(objset_t *os, void *tag)
622 dsl_dataset_disown(os->os_dsl_dataset, tag);
626 dmu_objset_evict_dbufs(objset_t *os)
631 mutex_enter(&os->os_lock);
632 dn = list_head(&os->os_dnodes);
635 * Skip dnodes without holds. We have to do this dance
636 * because dnode_add_ref() only works if there is already a
637 * hold. If the dnode has no holds, then it has no dbufs.
639 if (dnode_add_ref(dn, FTAG)) {
640 list_insert_after(&os->os_dnodes, dn, &dn_marker);
641 mutex_exit(&os->os_lock);
643 dnode_evict_dbufs(dn);
644 dnode_rele(dn, FTAG);
646 mutex_enter(&os->os_lock);
647 dn = list_next(&os->os_dnodes, &dn_marker);
648 list_remove(&os->os_dnodes, &dn_marker);
650 dn = list_next(&os->os_dnodes, dn);
653 mutex_exit(&os->os_lock);
655 if (DMU_USERUSED_DNODE(os) != NULL) {
656 dnode_evict_dbufs(DMU_GROUPUSED_DNODE(os));
657 dnode_evict_dbufs(DMU_USERUSED_DNODE(os));
659 dnode_evict_dbufs(DMU_META_DNODE(os));
663 * Objset eviction processing is split into into two pieces.
664 * The first marks the objset as evicting, evicts any dbufs that
665 * have a refcount of zero, and then queues up the objset for the
666 * second phase of eviction. Once os->os_dnodes has been cleared by
667 * dnode_buf_pageout()->dnode_destroy(), the second phase is executed.
668 * The second phase closes the special dnodes, dequeues the objset from
669 * the list of those undergoing eviction, and finally frees the objset.
671 * NOTE: Due to asynchronous eviction processing (invocation of
672 * dnode_buf_pageout()), it is possible for the meta dnode for the
673 * objset to have no holds even though os->os_dnodes is not empty.
676 dmu_objset_evict(objset_t *os)
678 dsl_dataset_t *ds = os->os_dsl_dataset;
680 for (int t = 0; t < TXG_SIZE; t++)
681 ASSERT(!dmu_objset_is_dirty(os, t));
684 dsl_prop_unregister_all(ds, os);
689 os->os_evicting = B_TRUE;
690 dmu_objset_evict_dbufs(os);
692 mutex_enter(&os->os_lock);
693 spa_evicting_os_register(os->os_spa, os);
694 if (list_is_empty(&os->os_dnodes)) {
695 mutex_exit(&os->os_lock);
696 dmu_objset_evict_done(os);
698 mutex_exit(&os->os_lock);
703 dmu_objset_evict_done(objset_t *os)
705 ASSERT3P(list_head(&os->os_dnodes), ==, NULL);
707 dnode_special_close(&os->os_meta_dnode);
708 if (DMU_USERUSED_DNODE(os)) {
709 dnode_special_close(&os->os_userused_dnode);
710 dnode_special_close(&os->os_groupused_dnode);
712 zil_free(os->os_zil);
714 VERIFY(arc_buf_remove_ref(os->os_phys_buf, &os->os_phys_buf));
717 * This is a barrier to prevent the objset from going away in
718 * dnode_move() until we can safely ensure that the objset is still in
719 * use. We consider the objset valid before the barrier and invalid
722 rw_enter(&os_lock, RW_READER);
725 mutex_destroy(&os->os_lock);
726 mutex_destroy(&os->os_obj_lock);
727 mutex_destroy(&os->os_user_ptr_lock);
728 spa_evicting_os_deregister(os->os_spa, os);
729 kmem_free(os, sizeof (objset_t));
733 dmu_objset_snap_cmtime(objset_t *os)
735 return (dsl_dir_snap_cmtime(os->os_dsl_dataset->ds_dir));
738 /* called from dsl for meta-objset */
740 dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
741 dmu_objset_type_t type, dmu_tx_t *tx)
746 ASSERT(dmu_tx_is_syncing(tx));
749 VERIFY0(dmu_objset_from_ds(ds, &os));
751 VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
753 mdn = DMU_META_DNODE(os);
755 dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
756 DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
759 * We don't want to have to increase the meta-dnode's nlevels
760 * later, because then we could do it in quescing context while
761 * we are also accessing it in open context.
763 * This precaution is not necessary for the MOS (ds == NULL),
764 * because the MOS is only updated in syncing context.
765 * This is most fortunate: the MOS is the only objset that
766 * needs to be synced multiple times as spa_sync() iterates
767 * to convergence, so minimizing its dn_nlevels matters.
773 * Determine the number of levels necessary for the meta-dnode
774 * to contain DN_MAX_OBJECT dnodes.
776 while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
777 (levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
778 DN_MAX_OBJECT * sizeof (dnode_phys_t))
781 mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
782 mdn->dn_nlevels = levels;
785 ASSERT(type != DMU_OST_NONE);
786 ASSERT(type != DMU_OST_ANY);
787 ASSERT(type < DMU_OST_NUMTYPES);
788 os->os_phys->os_type = type;
789 if (dmu_objset_userused_enabled(os)) {
790 os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
791 os->os_flags = os->os_phys->os_flags;
794 dsl_dataset_dirty(ds, tx);
799 typedef struct dmu_objset_create_arg {
800 const char *doca_name;
802 void (*doca_userfunc)(objset_t *os, void *arg,
803 cred_t *cr, dmu_tx_t *tx);
805 dmu_objset_type_t doca_type;
807 } dmu_objset_create_arg_t;
811 dmu_objset_create_check(void *arg, dmu_tx_t *tx)
813 dmu_objset_create_arg_t *doca = arg;
814 dsl_pool_t *dp = dmu_tx_pool(tx);
819 if (strchr(doca->doca_name, '@') != NULL)
820 return (SET_ERROR(EINVAL));
822 error = dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail);
826 dsl_dir_rele(pdd, FTAG);
827 return (SET_ERROR(EEXIST));
829 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
831 dsl_dir_rele(pdd, FTAG);
837 dmu_objset_create_sync(void *arg, dmu_tx_t *tx)
839 dmu_objset_create_arg_t *doca = arg;
840 dsl_pool_t *dp = dmu_tx_pool(tx);
848 VERIFY0(dsl_dir_hold(dp, doca->doca_name, FTAG, &pdd, &tail));
850 obj = dsl_dataset_create_sync(pdd, tail, NULL, doca->doca_flags,
851 doca->doca_cred, tx);
853 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
854 bp = dsl_dataset_get_blkptr(ds);
855 os = dmu_objset_create_impl(pdd->dd_pool->dp_spa,
856 ds, bp, doca->doca_type, tx);
858 if (doca->doca_userfunc != NULL) {
859 doca->doca_userfunc(os, doca->doca_userarg,
860 doca->doca_cred, tx);
863 spa_history_log_internal_ds(ds, "create", tx, "");
864 dsl_dataset_rele(ds, FTAG);
865 dsl_dir_rele(pdd, FTAG);
869 dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,
870 void (*func)(objset_t *os, void *arg, cred_t *cr, dmu_tx_t *tx), void *arg)
872 dmu_objset_create_arg_t doca;
874 doca.doca_name = name;
875 doca.doca_cred = CRED();
876 doca.doca_flags = flags;
877 doca.doca_userfunc = func;
878 doca.doca_userarg = arg;
879 doca.doca_type = type;
881 return (dsl_sync_task(name,
882 dmu_objset_create_check, dmu_objset_create_sync, &doca,
883 5, ZFS_SPACE_CHECK_NORMAL));
886 typedef struct dmu_objset_clone_arg {
887 const char *doca_clone;
888 const char *doca_origin;
890 } dmu_objset_clone_arg_t;
894 dmu_objset_clone_check(void *arg, dmu_tx_t *tx)
896 dmu_objset_clone_arg_t *doca = arg;
900 dsl_dataset_t *origin;
901 dsl_pool_t *dp = dmu_tx_pool(tx);
903 if (strchr(doca->doca_clone, '@') != NULL)
904 return (SET_ERROR(EINVAL));
906 error = dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail);
910 dsl_dir_rele(pdd, FTAG);
911 return (SET_ERROR(EEXIST));
914 error = dsl_fs_ss_limit_check(pdd, 1, ZFS_PROP_FILESYSTEM_LIMIT, NULL,
917 dsl_dir_rele(pdd, FTAG);
918 return (SET_ERROR(EDQUOT));
920 dsl_dir_rele(pdd, FTAG);
922 error = dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin);
926 /* You can only clone snapshots, not the head datasets. */
927 if (!origin->ds_is_snapshot) {
928 dsl_dataset_rele(origin, FTAG);
929 return (SET_ERROR(EINVAL));
931 dsl_dataset_rele(origin, FTAG);
937 dmu_objset_clone_sync(void *arg, dmu_tx_t *tx)
939 dmu_objset_clone_arg_t *doca = arg;
940 dsl_pool_t *dp = dmu_tx_pool(tx);
943 dsl_dataset_t *origin, *ds;
945 char namebuf[MAXNAMELEN];
947 VERIFY0(dsl_dir_hold(dp, doca->doca_clone, FTAG, &pdd, &tail));
948 VERIFY0(dsl_dataset_hold(dp, doca->doca_origin, FTAG, &origin));
950 obj = dsl_dataset_create_sync(pdd, tail, origin, 0,
951 doca->doca_cred, tx);
953 VERIFY0(dsl_dataset_hold_obj(pdd->dd_pool, obj, FTAG, &ds));
954 dsl_dataset_name(origin, namebuf);
955 spa_history_log_internal_ds(ds, "clone", tx,
956 "origin=%s (%llu)", namebuf, origin->ds_object);
957 dsl_dataset_rele(ds, FTAG);
958 dsl_dataset_rele(origin, FTAG);
959 dsl_dir_rele(pdd, FTAG);
963 dmu_objset_clone(const char *clone, const char *origin)
965 dmu_objset_clone_arg_t doca;
967 doca.doca_clone = clone;
968 doca.doca_origin = origin;
969 doca.doca_cred = CRED();
971 return (dsl_sync_task(clone,
972 dmu_objset_clone_check, dmu_objset_clone_sync, &doca,
973 5, ZFS_SPACE_CHECK_NORMAL));
977 dmu_objset_snapshot_one(const char *fsname, const char *snapname)
980 char *longsnap = kmem_asprintf("%s@%s", fsname, snapname);
981 nvlist_t *snaps = fnvlist_alloc();
983 fnvlist_add_boolean(snaps, longsnap);
985 err = dsl_dataset_snapshot(snaps, NULL, NULL);
991 dmu_objset_sync_dnodes(list_t *list, list_t *newlist, dmu_tx_t *tx)
995 while (dn = list_head(list)) {
996 ASSERT(dn->dn_object != DMU_META_DNODE_OBJECT);
997 ASSERT(dn->dn_dbuf->db_data_pending);
999 * Initialize dn_zio outside dnode_sync() because the
1000 * meta-dnode needs to set it ouside dnode_sync().
1002 dn->dn_zio = dn->dn_dbuf->db_data_pending->dr_zio;
1005 ASSERT3U(dn->dn_nlevels, <=, DN_MAX_LEVELS);
1006 list_remove(list, dn);
1009 (void) dnode_add_ref(dn, newlist);
1010 list_insert_tail(newlist, dn);
1019 dmu_objset_write_ready(zio_t *zio, arc_buf_t *abuf, void *arg)
1021 blkptr_t *bp = zio->io_bp;
1023 dnode_phys_t *dnp = &os->os_phys->os_meta_dnode;
1025 ASSERT(!BP_IS_EMBEDDED(bp));
1026 ASSERT3P(bp, ==, os->os_rootbp);
1027 ASSERT3U(BP_GET_TYPE(bp), ==, DMU_OT_OBJSET);
1028 ASSERT0(BP_GET_LEVEL(bp));
1031 * Update rootbp fill count: it should be the number of objects
1032 * allocated in the object set (not counting the "special"
1033 * objects that are stored in the objset_phys_t -- the meta
1034 * dnode and user/group accounting objects).
1037 for (int i = 0; i < dnp->dn_nblkptr; i++)
1038 bp->blk_fill += BP_GET_FILL(&dnp->dn_blkptr[i]);
1043 dmu_objset_write_done(zio_t *zio, arc_buf_t *abuf, void *arg)
1045 blkptr_t *bp = zio->io_bp;
1046 blkptr_t *bp_orig = &zio->io_bp_orig;
1049 if (zio->io_flags & ZIO_FLAG_IO_REWRITE) {
1050 ASSERT(BP_EQUAL(bp, bp_orig));
1052 dsl_dataset_t *ds = os->os_dsl_dataset;
1053 dmu_tx_t *tx = os->os_synctx;
1055 (void) dsl_dataset_block_kill(ds, bp_orig, tx, B_TRUE);
1056 dsl_dataset_block_born(ds, bp, tx);
1060 /* called from dsl */
1062 dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
1065 zbookmark_phys_t zb;
1069 list_t *newlist = NULL;
1070 dbuf_dirty_record_t *dr;
1072 dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
1074 ASSERT(dmu_tx_is_syncing(tx));
1075 /* XXX the write_done callback should really give us the tx... */
1078 if (os->os_dsl_dataset == NULL) {
1080 * This is the MOS. If we have upgraded,
1081 * spa_max_replication() could change, so reset
1084 os->os_copies = spa_max_replication(os->os_spa);
1088 * Create the root block IO
1090 SET_BOOKMARK(&zb, os->os_dsl_dataset ?
1091 os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
1092 ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
1093 arc_release(os->os_phys_buf, &os->os_phys_buf);
1095 dmu_write_policy(os, NULL, 0, 0, &zp);
1097 zio = arc_write(pio, os->os_spa, tx->tx_txg,
1098 os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os),
1099 DMU_OS_IS_L2COMPRESSIBLE(os), &zp, dmu_objset_write_ready,
1100 NULL, dmu_objset_write_done, os, ZIO_PRIORITY_ASYNC_WRITE,
1101 ZIO_FLAG_MUSTSUCCEED, &zb);
1104 * Sync special dnodes - the parent IO for the sync is the root block
1106 DMU_META_DNODE(os)->dn_zio = zio;
1107 dnode_sync(DMU_META_DNODE(os), tx);
1109 os->os_phys->os_flags = os->os_flags;
1111 if (DMU_USERUSED_DNODE(os) &&
1112 DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
1113 DMU_USERUSED_DNODE(os)->dn_zio = zio;
1114 dnode_sync(DMU_USERUSED_DNODE(os), tx);
1115 DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
1116 dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
1119 txgoff = tx->tx_txg & TXG_MASK;
1121 if (dmu_objset_userused_enabled(os)) {
1122 newlist = &os->os_synced_dnodes;
1124 * We must create the list here because it uses the
1125 * dn_dirty_link[] of this txg.
1127 list_create(newlist, sizeof (dnode_t),
1128 offsetof(dnode_t, dn_dirty_link[txgoff]));
1131 dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
1132 dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
1134 list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
1135 while (dr = list_head(list)) {
1136 ASSERT0(dr->dr_dbuf->db_level);
1137 list_remove(list, dr);
1139 zio_nowait(dr->dr_zio);
1142 * Free intent log blocks up to this tx.
1144 zil_sync(os->os_zil, tx);
1145 os->os_phys->os_zil_header = os->os_zil_header;
1150 dmu_objset_is_dirty(objset_t *os, uint64_t txg)
1152 return (!list_is_empty(&os->os_dirty_dnodes[txg & TXG_MASK]) ||
1153 !list_is_empty(&os->os_free_dnodes[txg & TXG_MASK]));
1156 static objset_used_cb_t *used_cbs[DMU_OST_NUMTYPES];
1159 dmu_objset_register_type(dmu_objset_type_t ost, objset_used_cb_t *cb)
1165 dmu_objset_userused_enabled(objset_t *os)
1167 return (spa_version(os->os_spa) >= SPA_VERSION_USERSPACE &&
1168 used_cbs[os->os_phys->os_type] != NULL &&
1169 DMU_USERUSED_DNODE(os) != NULL);
1173 do_userquota_update(objset_t *os, uint64_t used, uint64_t flags,
1174 uint64_t user, uint64_t group, boolean_t subtract, dmu_tx_t *tx)
1176 if ((flags & DNODE_FLAG_USERUSED_ACCOUNTED)) {
1177 int64_t delta = DNODE_SIZE + used;
1180 VERIFY3U(0, ==, zap_increment_int(os, DMU_USERUSED_OBJECT,
1182 VERIFY3U(0, ==, zap_increment_int(os, DMU_GROUPUSED_OBJECT,
1188 dmu_objset_do_userquota_updates(objset_t *os, dmu_tx_t *tx)
1191 list_t *list = &os->os_synced_dnodes;
1193 ASSERT(list_head(list) == NULL || dmu_objset_userused_enabled(os));
1195 while (dn = list_head(list)) {
1197 ASSERT(!DMU_OBJECT_IS_SPECIAL(dn->dn_object));
1198 ASSERT(dn->dn_phys->dn_type == DMU_OT_NONE ||
1199 dn->dn_phys->dn_flags &
1200 DNODE_FLAG_USERUSED_ACCOUNTED);
1202 /* Allocate the user/groupused objects if necessary. */
1203 if (DMU_USERUSED_DNODE(os)->dn_type == DMU_OT_NONE) {
1204 VERIFY(0 == zap_create_claim(os,
1205 DMU_USERUSED_OBJECT,
1206 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1207 VERIFY(0 == zap_create_claim(os,
1208 DMU_GROUPUSED_OBJECT,
1209 DMU_OT_USERGROUP_USED, DMU_OT_NONE, 0, tx));
1213 * We intentionally modify the zap object even if the
1214 * net delta is zero. Otherwise
1215 * the block of the zap obj could be shared between
1216 * datasets but need to be different between them after
1220 flags = dn->dn_id_flags;
1222 if (flags & DN_ID_OLD_EXIST) {
1223 do_userquota_update(os, dn->dn_oldused, dn->dn_oldflags,
1224 dn->dn_olduid, dn->dn_oldgid, B_TRUE, tx);
1226 if (flags & DN_ID_NEW_EXIST) {
1227 do_userquota_update(os, DN_USED_BYTES(dn->dn_phys),
1228 dn->dn_phys->dn_flags, dn->dn_newuid,
1229 dn->dn_newgid, B_FALSE, tx);
1232 mutex_enter(&dn->dn_mtx);
1234 dn->dn_oldflags = 0;
1235 if (dn->dn_id_flags & DN_ID_NEW_EXIST) {
1236 dn->dn_olduid = dn->dn_newuid;
1237 dn->dn_oldgid = dn->dn_newgid;
1238 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1239 if (dn->dn_bonuslen == 0)
1240 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1242 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1244 dn->dn_id_flags &= ~(DN_ID_NEW_EXIST);
1245 mutex_exit(&dn->dn_mtx);
1247 list_remove(list, dn);
1248 dnode_rele(dn, list);
1253 * Returns a pointer to data to find uid/gid from
1255 * If a dirty record for transaction group that is syncing can't
1256 * be found then NULL is returned. In the NULL case it is assumed
1257 * the uid/gid aren't changing.
1260 dmu_objset_userquota_find_data(dmu_buf_impl_t *db, dmu_tx_t *tx)
1262 dbuf_dirty_record_t *dr, **drp;
1265 if (db->db_dirtycnt == 0)
1266 return (db->db.db_data); /* Nothing is changing */
1268 for (drp = &db->db_last_dirty; (dr = *drp) != NULL; drp = &dr->dr_next)
1269 if (dr->dr_txg == tx->tx_txg)
1277 DB_DNODE_ENTER(dr->dr_dbuf);
1278 dn = DB_DNODE(dr->dr_dbuf);
1280 if (dn->dn_bonuslen == 0 &&
1281 dr->dr_dbuf->db_blkid == DMU_SPILL_BLKID)
1282 data = dr->dt.dl.dr_data->b_data;
1284 data = dr->dt.dl.dr_data;
1286 DB_DNODE_EXIT(dr->dr_dbuf);
1293 dmu_objset_userquota_get_ids(dnode_t *dn, boolean_t before, dmu_tx_t *tx)
1295 objset_t *os = dn->dn_objset;
1297 dmu_buf_impl_t *db = NULL;
1298 uint64_t *user = NULL;
1299 uint64_t *group = NULL;
1300 int flags = dn->dn_id_flags;
1302 boolean_t have_spill = B_FALSE;
1304 if (!dmu_objset_userused_enabled(dn->dn_objset))
1307 if (before && (flags & (DN_ID_CHKED_BONUS|DN_ID_OLD_EXIST|
1308 DN_ID_CHKED_SPILL)))
1311 if (before && dn->dn_bonuslen != 0)
1312 data = DN_BONUS(dn->dn_phys);
1313 else if (!before && dn->dn_bonuslen != 0) {
1316 mutex_enter(&db->db_mtx);
1317 data = dmu_objset_userquota_find_data(db, tx);
1319 data = DN_BONUS(dn->dn_phys);
1321 } else if (dn->dn_bonuslen == 0 && dn->dn_bonustype == DMU_OT_SA) {
1324 if (RW_WRITE_HELD(&dn->dn_struct_rwlock))
1325 rf |= DB_RF_HAVESTRUCT;
1326 error = dmu_spill_hold_by_dnode(dn,
1327 rf | DB_RF_MUST_SUCCEED,
1328 FTAG, (dmu_buf_t **)&db);
1330 mutex_enter(&db->db_mtx);
1331 data = (before) ? db->db.db_data :
1332 dmu_objset_userquota_find_data(db, tx);
1333 have_spill = B_TRUE;
1335 mutex_enter(&dn->dn_mtx);
1336 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1337 mutex_exit(&dn->dn_mtx);
1343 user = &dn->dn_olduid;
1344 group = &dn->dn_oldgid;
1346 user = &dn->dn_newuid;
1347 group = &dn->dn_newgid;
1351 * Must always call the callback in case the object
1352 * type has changed and that type isn't an object type to track
1354 error = used_cbs[os->os_phys->os_type](dn->dn_bonustype, data,
1358 * Preserve existing uid/gid when the callback can't determine
1359 * what the new uid/gid are and the callback returned EEXIST.
1360 * The EEXIST error tells us to just use the existing uid/gid.
1361 * If we don't know what the old values are then just assign
1362 * them to 0, since that is a new file being created.
1364 if (!before && data == NULL && error == EEXIST) {
1365 if (flags & DN_ID_OLD_EXIST) {
1366 dn->dn_newuid = dn->dn_olduid;
1367 dn->dn_newgid = dn->dn_oldgid;
1376 mutex_exit(&db->db_mtx);
1378 mutex_enter(&dn->dn_mtx);
1379 if (error == 0 && before)
1380 dn->dn_id_flags |= DN_ID_OLD_EXIST;
1381 if (error == 0 && !before)
1382 dn->dn_id_flags |= DN_ID_NEW_EXIST;
1385 dn->dn_id_flags |= DN_ID_CHKED_SPILL;
1387 dn->dn_id_flags |= DN_ID_CHKED_BONUS;
1389 mutex_exit(&dn->dn_mtx);
1391 dmu_buf_rele((dmu_buf_t *)db, FTAG);
1395 dmu_objset_userspace_present(objset_t *os)
1397 return (os->os_phys->os_flags &
1398 OBJSET_FLAG_USERACCOUNTING_COMPLETE);
1402 dmu_objset_userspace_upgrade(objset_t *os)
1407 if (dmu_objset_userspace_present(os))
1409 if (!dmu_objset_userused_enabled(os))
1410 return (SET_ERROR(ENOTSUP));
1411 if (dmu_objset_is_snapshot(os))
1412 return (SET_ERROR(EINVAL));
1415 * We simply need to mark every object dirty, so that it will be
1416 * synced out and now accounted. If this is called
1417 * concurrently, or if we already did some work before crashing,
1418 * that's fine, since we track each object's accounted state
1422 for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
1427 if (issig(JUSTLOOKING) && issig(FORREAL))
1428 return (SET_ERROR(EINTR));
1430 objerr = dmu_bonus_hold(os, obj, FTAG, &db);
1433 tx = dmu_tx_create(os);
1434 dmu_tx_hold_bonus(tx, obj);
1435 objerr = dmu_tx_assign(tx, TXG_WAIT);
1440 dmu_buf_will_dirty(db, tx);
1441 dmu_buf_rele(db, FTAG);
1445 os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
1446 txg_wait_synced(dmu_objset_pool(os), 0);
1451 dmu_objset_space(objset_t *os, uint64_t *refdbytesp, uint64_t *availbytesp,
1452 uint64_t *usedobjsp, uint64_t *availobjsp)
1454 dsl_dataset_space(os->os_dsl_dataset, refdbytesp, availbytesp,
1455 usedobjsp, availobjsp);
1459 dmu_objset_fsid_guid(objset_t *os)
1461 return (dsl_dataset_fsid_guid(os->os_dsl_dataset));
1465 dmu_objset_fast_stat(objset_t *os, dmu_objset_stats_t *stat)
1467 stat->dds_type = os->os_phys->os_type;
1468 if (os->os_dsl_dataset)
1469 dsl_dataset_fast_stat(os->os_dsl_dataset, stat);
1473 dmu_objset_stats(objset_t *os, nvlist_t *nv)
1475 ASSERT(os->os_dsl_dataset ||
1476 os->os_phys->os_type == DMU_OST_META);
1478 if (os->os_dsl_dataset != NULL)
1479 dsl_dataset_stats(os->os_dsl_dataset, nv);
1481 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_TYPE,
1482 os->os_phys->os_type);
1483 dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERACCOUNTING,
1484 dmu_objset_userspace_present(os));
1488 dmu_objset_is_snapshot(objset_t *os)
1490 if (os->os_dsl_dataset != NULL)
1491 return (os->os_dsl_dataset->ds_is_snapshot);
1497 dmu_snapshot_realname(objset_t *os, char *name, char *real, int maxlen,
1498 boolean_t *conflict)
1500 dsl_dataset_t *ds = os->os_dsl_dataset;
1503 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1504 return (SET_ERROR(ENOENT));
1506 return (zap_lookup_norm(ds->ds_dir->dd_pool->dp_meta_objset,
1507 dsl_dataset_phys(ds)->ds_snapnames_zapobj, name, 8, 1, &ignored,
1508 MT_FIRST, real, maxlen, conflict));
1512 dmu_snapshot_list_next(objset_t *os, int namelen, char *name,
1513 uint64_t *idp, uint64_t *offp, boolean_t *case_conflict)
1515 dsl_dataset_t *ds = os->os_dsl_dataset;
1516 zap_cursor_t cursor;
1517 zap_attribute_t attr;
1519 ASSERT(dsl_pool_config_held(dmu_objset_pool(os)));
1521 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj == 0)
1522 return (SET_ERROR(ENOENT));
1524 zap_cursor_init_serialized(&cursor,
1525 ds->ds_dir->dd_pool->dp_meta_objset,
1526 dsl_dataset_phys(ds)->ds_snapnames_zapobj, *offp);
1528 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1529 zap_cursor_fini(&cursor);
1530 return (SET_ERROR(ENOENT));
1533 if (strlen(attr.za_name) + 1 > namelen) {
1534 zap_cursor_fini(&cursor);
1535 return (SET_ERROR(ENAMETOOLONG));
1538 (void) strcpy(name, attr.za_name);
1540 *idp = attr.za_first_integer;
1542 *case_conflict = attr.za_normalization_conflict;
1543 zap_cursor_advance(&cursor);
1544 *offp = zap_cursor_serialize(&cursor);
1545 zap_cursor_fini(&cursor);
1551 dmu_dir_list_next(objset_t *os, int namelen, char *name,
1552 uint64_t *idp, uint64_t *offp)
1554 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1555 zap_cursor_t cursor;
1556 zap_attribute_t attr;
1558 /* there is no next dir on a snapshot! */
1559 if (os->os_dsl_dataset->ds_object !=
1560 dsl_dir_phys(dd)->dd_head_dataset_obj)
1561 return (SET_ERROR(ENOENT));
1563 zap_cursor_init_serialized(&cursor,
1564 dd->dd_pool->dp_meta_objset,
1565 dsl_dir_phys(dd)->dd_child_dir_zapobj, *offp);
1567 if (zap_cursor_retrieve(&cursor, &attr) != 0) {
1568 zap_cursor_fini(&cursor);
1569 return (SET_ERROR(ENOENT));
1572 if (strlen(attr.za_name) + 1 > namelen) {
1573 zap_cursor_fini(&cursor);
1574 return (SET_ERROR(ENAMETOOLONG));
1577 (void) strcpy(name, attr.za_name);
1579 *idp = attr.za_first_integer;
1580 zap_cursor_advance(&cursor);
1581 *offp = zap_cursor_serialize(&cursor);
1582 zap_cursor_fini(&cursor);
1587 typedef struct dmu_objset_find_ctx {
1591 int (*dc_func)(dsl_pool_t *, dsl_dataset_t *, void *);
1594 kmutex_t *dc_error_lock;
1596 } dmu_objset_find_ctx_t;
1599 dmu_objset_find_dp_impl(dmu_objset_find_ctx_t *dcp)
1601 dsl_pool_t *dp = dcp->dc_dp;
1602 dmu_objset_find_ctx_t *child_dcp;
1606 zap_attribute_t *attr;
1610 /* don't process if there already was an error */
1611 if (*dcp->dc_error != 0)
1614 err = dsl_dir_hold_obj(dp, dcp->dc_ddobj, NULL, FTAG, &dd);
1618 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1619 if (dd->dd_myname[0] == '$') {
1620 dsl_dir_rele(dd, FTAG);
1624 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1625 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1628 * Iterate over all children.
1630 if (dcp->dc_flags & DS_FIND_CHILDREN) {
1631 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1632 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1633 zap_cursor_retrieve(&zc, attr) == 0;
1634 (void) zap_cursor_advance(&zc)) {
1635 ASSERT3U(attr->za_integer_length, ==,
1637 ASSERT3U(attr->za_num_integers, ==, 1);
1639 child_dcp = kmem_alloc(sizeof (*child_dcp), KM_SLEEP);
1641 child_dcp->dc_ddobj = attr->za_first_integer;
1642 if (dcp->dc_tq != NULL)
1643 (void) taskq_dispatch(dcp->dc_tq,
1644 dmu_objset_find_dp_cb, child_dcp, TQ_SLEEP);
1646 dmu_objset_find_dp_impl(child_dcp);
1648 zap_cursor_fini(&zc);
1652 * Iterate over all snapshots.
1654 if (dcp->dc_flags & DS_FIND_SNAPSHOTS) {
1656 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1661 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1662 dsl_dataset_rele(ds, FTAG);
1664 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1665 zap_cursor_retrieve(&zc, attr) == 0;
1666 (void) zap_cursor_advance(&zc)) {
1667 ASSERT3U(attr->za_integer_length, ==,
1669 ASSERT3U(attr->za_num_integers, ==, 1);
1671 err = dsl_dataset_hold_obj(dp,
1672 attr->za_first_integer, FTAG, &ds);
1675 err = dcp->dc_func(dp, ds, dcp->dc_arg);
1676 dsl_dataset_rele(ds, FTAG);
1680 zap_cursor_fini(&zc);
1684 dsl_dir_rele(dd, FTAG);
1685 kmem_free(attr, sizeof (zap_attribute_t));
1693 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1696 err = dcp->dc_func(dp, ds, dcp->dc_arg);
1697 dsl_dataset_rele(ds, FTAG);
1701 mutex_enter(dcp->dc_error_lock);
1702 /* only keep first error */
1703 if (*dcp->dc_error == 0)
1704 *dcp->dc_error = err;
1705 mutex_exit(dcp->dc_error_lock);
1708 kmem_free(dcp, sizeof (*dcp));
1712 dmu_objset_find_dp_cb(void *arg)
1714 dmu_objset_find_ctx_t *dcp = arg;
1715 dsl_pool_t *dp = dcp->dc_dp;
1718 * We need to get a pool_config_lock here, as there are several
1719 * asssert(pool_config_held) down the stack. Getting a lock via
1720 * dsl_pool_config_enter is risky, as it might be stalled by a
1721 * pending writer. This would deadlock, as the write lock can
1722 * only be granted when our parent thread gives up the lock.
1723 * The _prio interface gives us priority over a pending writer.
1725 dsl_pool_config_enter_prio(dp, FTAG);
1727 dmu_objset_find_dp_impl(dcp);
1729 dsl_pool_config_exit(dp, FTAG);
1733 * Find objsets under and including ddobj, call func(ds) on each.
1734 * The order for the enumeration is completely undefined.
1735 * func is called with dsl_pool_config held.
1738 dmu_objset_find_dp(dsl_pool_t *dp, uint64_t ddobj,
1739 int func(dsl_pool_t *, dsl_dataset_t *, void *), void *arg, int flags)
1744 dmu_objset_find_ctx_t *dcp;
1747 mutex_init(&err_lock, NULL, MUTEX_DEFAULT, NULL);
1748 dcp = kmem_alloc(sizeof (*dcp), KM_SLEEP);
1751 dcp->dc_ddobj = ddobj;
1752 dcp->dc_func = func;
1754 dcp->dc_flags = flags;
1755 dcp->dc_error_lock = &err_lock;
1756 dcp->dc_error = &error;
1758 if ((flags & DS_FIND_SERIALIZE) || dsl_pool_config_held_writer(dp)) {
1760 * In case a write lock is held we can't make use of
1761 * parallelism, as down the stack of the worker threads
1762 * the lock is asserted via dsl_pool_config_held.
1763 * In case of a read lock this is solved by getting a read
1764 * lock in each worker thread, which isn't possible in case
1765 * of a writer lock. So we fall back to the synchronous path
1767 * In the future it might be possible to get some magic into
1768 * dsl_pool_config_held in a way that it returns true for
1769 * the worker threads so that a single lock held from this
1770 * thread suffices. For now, stay single threaded.
1772 dmu_objset_find_dp_impl(dcp);
1777 ntasks = dmu_find_threads;
1779 ntasks = vdev_count_leaves(dp->dp_spa) * 4;
1780 tq = taskq_create("dmu_objset_find", ntasks, minclsyspri, ntasks,
1783 kmem_free(dcp, sizeof (*dcp));
1784 return (SET_ERROR(ENOMEM));
1788 /* dcp will be freed by task */
1789 (void) taskq_dispatch(tq, dmu_objset_find_dp_cb, dcp, TQ_SLEEP);
1792 * PORTING: this code relies on the property of taskq_wait to wait
1793 * until no more tasks are queued and no more tasks are active. As
1794 * we always queue new tasks from within other tasks, task_wait
1795 * reliably waits for the full recursion to finish, even though we
1796 * enqueue new tasks after taskq_wait has been called.
1797 * On platforms other than illumos, taskq_wait may not have this
1802 mutex_destroy(&err_lock);
1808 * Find all objsets under name, and for each, call 'func(child_name, arg)'.
1809 * The dp_config_rwlock must not be held when this is called, and it
1810 * will not be held when the callback is called.
1811 * Therefore this function should only be used when the pool is not changing
1812 * (e.g. in syncing context), or the callback can deal with the possible races.
1815 dmu_objset_find_impl(spa_t *spa, const char *name,
1816 int func(const char *, void *), void *arg, int flags)
1819 dsl_pool_t *dp = spa_get_dsl(spa);
1822 zap_attribute_t *attr;
1827 dsl_pool_config_enter(dp, FTAG);
1829 err = dsl_dir_hold(dp, name, FTAG, &dd, NULL);
1831 dsl_pool_config_exit(dp, FTAG);
1835 /* Don't visit hidden ($MOS & $ORIGIN) objsets. */
1836 if (dd->dd_myname[0] == '$') {
1837 dsl_dir_rele(dd, FTAG);
1838 dsl_pool_config_exit(dp, FTAG);
1842 thisobj = dsl_dir_phys(dd)->dd_head_dataset_obj;
1843 attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
1846 * Iterate over all children.
1848 if (flags & DS_FIND_CHILDREN) {
1849 for (zap_cursor_init(&zc, dp->dp_meta_objset,
1850 dsl_dir_phys(dd)->dd_child_dir_zapobj);
1851 zap_cursor_retrieve(&zc, attr) == 0;
1852 (void) zap_cursor_advance(&zc)) {
1853 ASSERT3U(attr->za_integer_length, ==,
1855 ASSERT3U(attr->za_num_integers, ==, 1);
1857 child = kmem_asprintf("%s/%s", name, attr->za_name);
1858 dsl_pool_config_exit(dp, FTAG);
1859 err = dmu_objset_find_impl(spa, child,
1861 dsl_pool_config_enter(dp, FTAG);
1866 zap_cursor_fini(&zc);
1869 dsl_dir_rele(dd, FTAG);
1870 dsl_pool_config_exit(dp, FTAG);
1871 kmem_free(attr, sizeof (zap_attribute_t));
1877 * Iterate over all snapshots.
1879 if (flags & DS_FIND_SNAPSHOTS) {
1880 err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
1885 snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
1886 dsl_dataset_rele(ds, FTAG);
1888 for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
1889 zap_cursor_retrieve(&zc, attr) == 0;
1890 (void) zap_cursor_advance(&zc)) {
1891 ASSERT3U(attr->za_integer_length, ==,
1893 ASSERT3U(attr->za_num_integers, ==, 1);
1895 child = kmem_asprintf("%s@%s",
1896 name, attr->za_name);
1897 dsl_pool_config_exit(dp, FTAG);
1898 err = func(child, arg);
1899 dsl_pool_config_enter(dp, FTAG);
1904 zap_cursor_fini(&zc);
1908 dsl_dir_rele(dd, FTAG);
1909 kmem_free(attr, sizeof (zap_attribute_t));
1910 dsl_pool_config_exit(dp, FTAG);
1915 /* Apply to self. */
1916 return (func(name, arg));
1920 * See comment above dmu_objset_find_impl().
1923 dmu_objset_find(char *name, int func(const char *, void *), void *arg,
1929 error = spa_open(name, &spa, FTAG);
1932 error = dmu_objset_find_impl(spa, name, func, arg, flags);
1933 spa_close(spa, FTAG);
1938 dmu_objset_set_user(objset_t *os, void *user_ptr)
1940 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1941 os->os_user_ptr = user_ptr;
1945 dmu_objset_get_user(objset_t *os)
1947 ASSERT(MUTEX_HELD(&os->os_user_ptr_lock));
1948 return (os->os_user_ptr);
1952 * Determine name of filesystem, given name of snapshot.
1953 * buf must be at least MAXNAMELEN bytes
1956 dmu_fsname(const char *snapname, char *buf)
1958 char *atp = strchr(snapname, '@');
1960 return (SET_ERROR(EINVAL));
1961 if (atp - snapname >= MAXNAMELEN)
1962 return (SET_ERROR(ENAMETOOLONG));
1963 (void) strlcpy(buf, snapname, atp - snapname + 1);