1/*
2 * CDDL HEADER START
3 *
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.
7 *
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.
12 *
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]
18 *
19 * CDDL HEADER END
20 */
21
22/*
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (c) 2011, 2017 by Delphix. All rights reserved.
25 * Copyright (c) 2014, Joyent, Inc. All rights reserved.
26 * Copyright (c) 2014 RackTop Systems.
27 * Copyright (c) 2014 Spectra Logic Corporation, All rights reserved.
28 * Copyright (c) 2014 Integros [integros.com]
29 * Copyright 2016, OmniTI Computer Consulting, Inc. All rights reserved.
30 * Copyright 2017 Nexenta Systems, Inc.
31 */
32
33#include <sys/dmu_objset.h>
34#include <sys/dsl_dataset.h>
35#include <sys/dsl_dir.h>
36#include <sys/dsl_prop.h>
37#include <sys/dsl_synctask.h>
38#include <sys/dmu_traverse.h>
39#include <sys/dmu_impl.h>
40#include <sys/dmu_tx.h>
41#include <sys/arc.h>
42#include <sys/zio.h>
43#include <sys/zap.h>
44#include <sys/zfeature.h>
45#include <sys/unique.h>
46#include <sys/zfs_context.h>
47#include <sys/zfs_ioctl.h>
48#include <sys/spa.h>
49#include <sys/spa_impl.h>
50#include <sys/vdev.h>
51#include <sys/zfs_znode.h>
52#include <sys/zfs_onexit.h>
53#include <sys/zvol.h>
54#include <sys/dsl_scan.h>
55#include <sys/dsl_deadlist.h>
56#include <sys/dsl_destroy.h>
57#include <sys/dsl_userhold.h>
58#include <sys/dsl_bookmark.h>
59#include <sys/dmu_recv.h>
60#include <sys/zio_checksum.h>
61#include <sys/zio_compress.h>
62#include <zfs_fletcher.h>
63
64/*
65 * The SPA supports block sizes up to 16MB.  However, very large blocks
66 * can have an impact on i/o latency (e.g. tying up a spinning disk for
67 * ~300ms), and also potentially on the memory allocator.  Therefore,
68 * we do not allow the recordsize to be set larger than zfs_max_recordsize
69 * (default 1MB).  Larger blocks can be created by changing this tunable,
70 * and pools with larger blocks can always be imported and used, regardless
71 * of this setting.
72 */
73int zfs_max_recordsize = 1 * 1024 * 1024;
74
75#define	SWITCH64(x, y) \
76	{ \
77		uint64_t __tmp = (x); \
78		(x) = (y); \
79		(y) = __tmp; \
80	}
81
82#define	DS_REF_MAX	(1ULL << 62)
83
84extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
85
86static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds,
87    uint64_t obj, dmu_tx_t *tx);
88static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds,
89    dmu_tx_t *tx);
90
91extern int spa_asize_inflation;
92
93static zil_header_t zero_zil;
94
95/*
96 * Figure out how much of this delta should be propogated to the dsl_dir
97 * layer.  If there's a refreservation, that space has already been
98 * partially accounted for in our ancestors.
99 */
100static int64_t
101parent_delta(dsl_dataset_t *ds, int64_t delta)
102{
103	dsl_dataset_phys_t *ds_phys;
104	uint64_t old_bytes, new_bytes;
105
106	if (ds->ds_reserved == 0)
107		return (delta);
108
109	ds_phys = dsl_dataset_phys(ds);
110	old_bytes = MAX(ds_phys->ds_unique_bytes, ds->ds_reserved);
111	new_bytes = MAX(ds_phys->ds_unique_bytes + delta, ds->ds_reserved);
112
113	ASSERT3U(ABS((int64_t)(new_bytes - old_bytes)), <=, ABS(delta));
114	return (new_bytes - old_bytes);
115}
116
117void
118dsl_dataset_block_born(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx)
119{
120	int used = bp_get_dsize_sync(tx->tx_pool->dp_spa, bp);
121	int compressed = BP_GET_PSIZE(bp);
122	int uncompressed = BP_GET_UCSIZE(bp);
123	int64_t delta;
124
125	dprintf_bp(bp, "ds=%p", ds);
126
127	ASSERT(dmu_tx_is_syncing(tx));
128	/* It could have been compressed away to nothing */
129	if (BP_IS_HOLE(bp))
130		return;
131	ASSERT(BP_GET_TYPE(bp) != DMU_OT_NONE);
132	ASSERT(DMU_OT_IS_VALID(BP_GET_TYPE(bp)));
133	if (ds == NULL) {
134		dsl_pool_mos_diduse_space(tx->tx_pool,
135		    used, compressed, uncompressed);
136		return;
137	}
138
139	ASSERT3U(bp->blk_birth, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
140	dmu_buf_will_dirty(ds->ds_dbuf, tx);
141	mutex_enter(&ds->ds_lock);
142	delta = parent_delta(ds, used);
143	dsl_dataset_phys(ds)->ds_referenced_bytes += used;
144	dsl_dataset_phys(ds)->ds_compressed_bytes += compressed;
145	dsl_dataset_phys(ds)->ds_uncompressed_bytes += uncompressed;
146	dsl_dataset_phys(ds)->ds_unique_bytes += used;
147
148	if (BP_GET_LSIZE(bp) > SPA_OLD_MAXBLOCKSIZE) {
149		ds->ds_feature_activation_needed[SPA_FEATURE_LARGE_BLOCKS] =
150		    B_TRUE;
151	}
152
153	spa_feature_t f = zio_checksum_to_feature(BP_GET_CHECKSUM(bp));
154	if (f != SPA_FEATURE_NONE)
155		ds->ds_feature_activation_needed[f] = B_TRUE;
156
157	mutex_exit(&ds->ds_lock);
158	dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD, delta,
159	    compressed, uncompressed, tx);
160	dsl_dir_transfer_space(ds->ds_dir, used - delta,
161	    DD_USED_REFRSRV, DD_USED_HEAD, tx);
162}
163
164/*
165 * Called when the specified segment has been remapped, and is thus no
166 * longer referenced in the head dataset.  The vdev must be indirect.
167 *
168 * If the segment is referenced by a snapshot, put it on the remap deadlist.
169 * Otherwise, add this segment to the obsolete spacemap.
170 */
171void
172dsl_dataset_block_remapped(dsl_dataset_t *ds, uint64_t vdev, uint64_t offset,
173    uint64_t size, uint64_t birth, dmu_tx_t *tx)
174{
175	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
176
177	ASSERT(dmu_tx_is_syncing(tx));
178	ASSERT(birth <= tx->tx_txg);
179	ASSERT(!ds->ds_is_snapshot);
180
181	if (birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
182		spa_vdev_indirect_mark_obsolete(spa, vdev, offset, size, tx);
183	} else {
184		blkptr_t fakebp;
185		dva_t *dva = &fakebp.blk_dva[0];
186
187		ASSERT(ds != NULL);
188
189		mutex_enter(&ds->ds_remap_deadlist_lock);
190		if (!dsl_dataset_remap_deadlist_exists(ds)) {
191			dsl_dataset_create_remap_deadlist(ds, tx);
192		}
193		mutex_exit(&ds->ds_remap_deadlist_lock);
194
195		BP_ZERO(&fakebp);
196		fakebp.blk_birth = birth;
197		DVA_SET_VDEV(dva, vdev);
198		DVA_SET_OFFSET(dva, offset);
199		DVA_SET_ASIZE(dva, size);
200
201		dsl_deadlist_insert(&ds->ds_remap_deadlist, &fakebp, tx);
202	}
203}
204
205int
206dsl_dataset_block_kill(dsl_dataset_t *ds, const blkptr_t *bp, dmu_tx_t *tx,
207    boolean_t async)
208{
209	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
210
211	int used = bp_get_dsize_sync(spa, bp);
212	int compressed = BP_GET_PSIZE(bp);
213	int uncompressed = BP_GET_UCSIZE(bp);
214
215	if (BP_IS_HOLE(bp))
216		return (0);
217
218	ASSERT(dmu_tx_is_syncing(tx));
219	ASSERT(bp->blk_birth <= tx->tx_txg);
220
221	if (ds == NULL) {
222		dsl_free(tx->tx_pool, tx->tx_txg, bp);
223		dsl_pool_mos_diduse_space(tx->tx_pool,
224		    -used, -compressed, -uncompressed);
225		return (used);
226	}
227	ASSERT3P(tx->tx_pool, ==, ds->ds_dir->dd_pool);
228
229	ASSERT(!ds->ds_is_snapshot);
230	dmu_buf_will_dirty(ds->ds_dbuf, tx);
231
232	if (bp->blk_birth > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
233		int64_t delta;
234
235		dprintf_bp(bp, "freeing ds=%llu", ds->ds_object);
236		dsl_free(tx->tx_pool, tx->tx_txg, bp);
237
238		mutex_enter(&ds->ds_lock);
239		ASSERT(dsl_dataset_phys(ds)->ds_unique_bytes >= used ||
240		    !DS_UNIQUE_IS_ACCURATE(ds));
241		delta = parent_delta(ds, -used);
242		dsl_dataset_phys(ds)->ds_unique_bytes -= used;
243		mutex_exit(&ds->ds_lock);
244		dsl_dir_diduse_space(ds->ds_dir, DD_USED_HEAD,
245		    delta, -compressed, -uncompressed, tx);
246		dsl_dir_transfer_space(ds->ds_dir, -used - delta,
247		    DD_USED_REFRSRV, DD_USED_HEAD, tx);
248	} else {
249		dprintf_bp(bp, "putting on dead list: %s", "");
250		if (async) {
251			/*
252			 * We are here as part of zio's write done callback,
253			 * which means we're a zio interrupt thread.  We can't
254			 * call dsl_deadlist_insert() now because it may block
255			 * waiting for I/O.  Instead, put bp on the deferred
256			 * queue and let dsl_pool_sync() finish the job.
257			 */
258			bplist_append(&ds->ds_pending_deadlist, bp);
259		} else {
260			dsl_deadlist_insert(&ds->ds_deadlist, bp, tx);
261		}
262		ASSERT3U(ds->ds_prev->ds_object, ==,
263		    dsl_dataset_phys(ds)->ds_prev_snap_obj);
264		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_num_children > 0);
265		/* if (bp->blk_birth > prev prev snap txg) prev unique += bs */
266		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
267		    ds->ds_object && bp->blk_birth >
268		    dsl_dataset_phys(ds->ds_prev)->ds_prev_snap_txg) {
269			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
270			mutex_enter(&ds->ds_prev->ds_lock);
271			dsl_dataset_phys(ds->ds_prev)->ds_unique_bytes += used;
272			mutex_exit(&ds->ds_prev->ds_lock);
273		}
274		if (bp->blk_birth > ds->ds_dir->dd_origin_txg) {
275			dsl_dir_transfer_space(ds->ds_dir, used,
276			    DD_USED_HEAD, DD_USED_SNAP, tx);
277		}
278	}
279	mutex_enter(&ds->ds_lock);
280	ASSERT3U(dsl_dataset_phys(ds)->ds_referenced_bytes, >=, used);
281	dsl_dataset_phys(ds)->ds_referenced_bytes -= used;
282	ASSERT3U(dsl_dataset_phys(ds)->ds_compressed_bytes, >=, compressed);
283	dsl_dataset_phys(ds)->ds_compressed_bytes -= compressed;
284	ASSERT3U(dsl_dataset_phys(ds)->ds_uncompressed_bytes, >=, uncompressed);
285	dsl_dataset_phys(ds)->ds_uncompressed_bytes -= uncompressed;
286	mutex_exit(&ds->ds_lock);
287
288	return (used);
289}
290
291/*
292 * We have to release the fsid syncronously or we risk that a subsequent
293 * mount of the same dataset will fail to unique_insert the fsid.  This
294 * failure would manifest itself as the fsid of this dataset changing
295 * between mounts which makes NFS clients quite unhappy.
296 */
297static void
298dsl_dataset_evict_sync(void *dbu)
299{
300	dsl_dataset_t *ds = dbu;
301
302	ASSERT(ds->ds_owner == NULL);
303
304	unique_remove(ds->ds_fsid_guid);
305}
306
307static void
308dsl_dataset_evict_async(void *dbu)
309{
310	dsl_dataset_t *ds = dbu;
311
312	ASSERT(ds->ds_owner == NULL);
313
314	ds->ds_dbuf = NULL;
315
316	if (ds->ds_objset != NULL)
317		dmu_objset_evict(ds->ds_objset);
318
319	if (ds->ds_prev) {
320		dsl_dataset_rele(ds->ds_prev, ds);
321		ds->ds_prev = NULL;
322	}
323
324	bplist_destroy(&ds->ds_pending_deadlist);
325	if (dsl_deadlist_is_open(&ds->ds_deadlist))
326		dsl_deadlist_close(&ds->ds_deadlist);
327	if (dsl_deadlist_is_open(&ds->ds_remap_deadlist))
328		dsl_deadlist_close(&ds->ds_remap_deadlist);
329	if (ds->ds_dir)
330		dsl_dir_async_rele(ds->ds_dir, ds);
331
332	ASSERT(!list_link_active(&ds->ds_synced_link));
333
334	list_destroy(&ds->ds_prop_cbs);
335	mutex_destroy(&ds->ds_lock);
336	mutex_destroy(&ds->ds_opening_lock);
337	mutex_destroy(&ds->ds_sendstream_lock);
338	mutex_destroy(&ds->ds_remap_deadlist_lock);
339	zfs_refcount_destroy(&ds->ds_longholds);
340	rrw_destroy(&ds->ds_bp_rwlock);
341
342	kmem_free(ds, sizeof (dsl_dataset_t));
343}
344
345int
346dsl_dataset_get_snapname(dsl_dataset_t *ds)
347{
348	dsl_dataset_phys_t *headphys;
349	int err;
350	dmu_buf_t *headdbuf;
351	dsl_pool_t *dp = ds->ds_dir->dd_pool;
352	objset_t *mos = dp->dp_meta_objset;
353
354	if (ds->ds_snapname[0])
355		return (0);
356	if (dsl_dataset_phys(ds)->ds_next_snap_obj == 0)
357		return (0);
358
359	err = dmu_bonus_hold(mos, dsl_dir_phys(ds->ds_dir)->dd_head_dataset_obj,
360	    FTAG, &headdbuf);
361	if (err != 0)
362		return (err);
363	headphys = headdbuf->db_data;
364	err = zap_value_search(dp->dp_meta_objset,
365	    headphys->ds_snapnames_zapobj, ds->ds_object, 0, ds->ds_snapname);
366	dmu_buf_rele(headdbuf, FTAG);
367	return (err);
368}
369
370int
371dsl_dataset_snap_lookup(dsl_dataset_t *ds, const char *name, uint64_t *value)
372{
373	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
374	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
375	matchtype_t mt = 0;
376	int err;
377
378	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
379		mt = MT_NORMALIZE;
380
381	err = zap_lookup_norm(mos, snapobj, name, 8, 1,
382	    value, mt, NULL, 0, NULL);
383	if (err == ENOTSUP && (mt & MT_NORMALIZE))
384		err = zap_lookup(mos, snapobj, name, 8, 1, value);
385	return (err);
386}
387
388int
389dsl_dataset_snap_remove(dsl_dataset_t *ds, const char *name, dmu_tx_t *tx,
390    boolean_t adj_cnt)
391{
392	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
393	uint64_t snapobj = dsl_dataset_phys(ds)->ds_snapnames_zapobj;
394	matchtype_t mt = 0;
395	int err;
396
397	dsl_dir_snap_cmtime_update(ds->ds_dir);
398
399	if (dsl_dataset_phys(ds)->ds_flags & DS_FLAG_CI_DATASET)
400		mt = MT_NORMALIZE;
401
402	err = zap_remove_norm(mos, snapobj, name, mt, tx);
403	if (err == ENOTSUP && (mt & MT_NORMALIZE))
404		err = zap_remove(mos, snapobj, name, tx);
405
406	if (err == 0 && adj_cnt)
407		dsl_fs_ss_count_adjust(ds->ds_dir, -1,
408		    DD_FIELD_SNAPSHOT_COUNT, tx);
409
410	return (err);
411}
412
413boolean_t
414dsl_dataset_try_add_ref(dsl_pool_t *dp, dsl_dataset_t *ds, void *tag)
415{
416	dmu_buf_t *dbuf = ds->ds_dbuf;
417	boolean_t result = B_FALSE;
418
419	if (dbuf != NULL && dmu_buf_try_add_ref(dbuf, dp->dp_meta_objset,
420	    ds->ds_object, DMU_BONUS_BLKID, tag)) {
421
422		if (ds == dmu_buf_get_user(dbuf))
423			result = B_TRUE;
424		else
425			dmu_buf_rele(dbuf, tag);
426	}
427
428	return (result);
429}
430
431int
432dsl_dataset_hold_obj(dsl_pool_t *dp, uint64_t dsobj, void *tag,
433    dsl_dataset_t **dsp)
434{
435	objset_t *mos = dp->dp_meta_objset;
436	dmu_buf_t *dbuf;
437	dsl_dataset_t *ds;
438	int err;
439	dmu_object_info_t doi;
440
441	ASSERT(dsl_pool_config_held(dp));
442
443	err = dmu_bonus_hold(mos, dsobj, tag, &dbuf);
444	if (err != 0)
445		return (err);
446
447	/* Make sure dsobj has the correct object type. */
448	dmu_object_info_from_db(dbuf, &doi);
449	if (doi.doi_bonus_type != DMU_OT_DSL_DATASET) {
450		dmu_buf_rele(dbuf, tag);
451		return (SET_ERROR(EINVAL));
452	}
453
454	ds = dmu_buf_get_user(dbuf);
455	if (ds == NULL) {
456		dsl_dataset_t *winner = NULL;
457
458		ds = kmem_zalloc(sizeof (dsl_dataset_t), KM_SLEEP);
459		ds->ds_dbuf = dbuf;
460		ds->ds_object = dsobj;
461		ds->ds_is_snapshot = dsl_dataset_phys(ds)->ds_num_children != 0;
462
463		err = dsl_dir_hold_obj(dp, dsl_dataset_phys(ds)->ds_dir_obj,
464		    NULL, ds, &ds->ds_dir);
465		if (err != 0) {
466			kmem_free(ds, sizeof (dsl_dataset_t));
467			dmu_buf_rele(dbuf, tag);
468			return (err);
469		}
470
471		mutex_init(&ds->ds_lock, NULL, MUTEX_DEFAULT, NULL);
472		mutex_init(&ds->ds_opening_lock, NULL, MUTEX_DEFAULT, NULL);
473		mutex_init(&ds->ds_sendstream_lock, NULL, MUTEX_DEFAULT, NULL);
474		mutex_init(&ds->ds_remap_deadlist_lock,
475		    NULL, MUTEX_DEFAULT, NULL);
476		rrw_init(&ds->ds_bp_rwlock, B_FALSE);
477		zfs_refcount_create(&ds->ds_longholds);
478
479		bplist_create(&ds->ds_pending_deadlist);
480
481		list_create(&ds->ds_sendstreams, sizeof (dmu_sendarg_t),
482		    offsetof(dmu_sendarg_t, dsa_link));
483
484		list_create(&ds->ds_prop_cbs, sizeof (dsl_prop_cb_record_t),
485		    offsetof(dsl_prop_cb_record_t, cbr_ds_node));
486
487		if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
488			for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
489				if (!(spa_feature_table[f].fi_flags &
490				    ZFEATURE_FLAG_PER_DATASET))
491					continue;
492				err = zap_contains(mos, dsobj,
493				    spa_feature_table[f].fi_guid);
494				if (err == 0) {
495					ds->ds_feature_inuse[f] = B_TRUE;
496				} else {
497					ASSERT3U(err, ==, ENOENT);
498					err = 0;
499				}
500			}
501		}
502
503		if (!ds->ds_is_snapshot) {
504			ds->ds_snapname[0] = '\0';
505			if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
506				err = dsl_dataset_hold_obj(dp,
507				    dsl_dataset_phys(ds)->ds_prev_snap_obj,
508				    ds, &ds->ds_prev);
509			}
510			if (doi.doi_type == DMU_OTN_ZAP_METADATA) {
511				int zaperr = zap_lookup(mos, ds->ds_object,
512				    DS_FIELD_BOOKMARK_NAMES,
513				    sizeof (ds->ds_bookmarks), 1,
514				    &ds->ds_bookmarks);
515				if (zaperr != ENOENT)
516					VERIFY0(zaperr);
517			}
518		} else {
519			if (zfs_flags & ZFS_DEBUG_SNAPNAMES)
520				err = dsl_dataset_get_snapname(ds);
521			if (err == 0 &&
522			    dsl_dataset_phys(ds)->ds_userrefs_obj != 0) {
523				err = zap_count(
524				    ds->ds_dir->dd_pool->dp_meta_objset,
525				    dsl_dataset_phys(ds)->ds_userrefs_obj,
526				    &ds->ds_userrefs);
527			}
528		}
529
530		if (err == 0 && !ds->ds_is_snapshot) {
531			err = dsl_prop_get_int_ds(ds,
532			    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
533			    &ds->ds_reserved);
534			if (err == 0) {
535				err = dsl_prop_get_int_ds(ds,
536				    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
537				    &ds->ds_quota);
538			}
539		} else {
540			ds->ds_reserved = ds->ds_quota = 0;
541		}
542
543		if (err == 0 && ds->ds_dir->dd_crypto_obj != 0 &&
544		    ds->ds_is_snapshot &&
545		    zap_contains(mos, dsobj, DS_FIELD_IVSET_GUID) != 0) {
546			dp->dp_spa->spa_errata =
547			    ZPOOL_ERRATA_ZOL_8308_ENCRYPTION;
548		}
549
550		dsl_deadlist_open(&ds->ds_deadlist,
551		    mos, dsl_dataset_phys(ds)->ds_deadlist_obj);
552		uint64_t remap_deadlist_obj =
553		    dsl_dataset_get_remap_deadlist_object(ds);
554		if (remap_deadlist_obj != 0) {
555			dsl_deadlist_open(&ds->ds_remap_deadlist, mos,
556			    remap_deadlist_obj);
557		}
558
559		dmu_buf_init_user(&ds->ds_dbu, dsl_dataset_evict_sync,
560		    dsl_dataset_evict_async, &ds->ds_dbuf);
561		if (err == 0)
562			winner = dmu_buf_set_user_ie(dbuf, &ds->ds_dbu);
563
564		if (err != 0 || winner != NULL) {
565			bplist_destroy(&ds->ds_pending_deadlist);
566			dsl_deadlist_close(&ds->ds_deadlist);
567			if (dsl_deadlist_is_open(&ds->ds_remap_deadlist))
568				dsl_deadlist_close(&ds->ds_remap_deadlist);
569			if (ds->ds_prev)
570				dsl_dataset_rele(ds->ds_prev, ds);
571			dsl_dir_rele(ds->ds_dir, ds);
572			list_destroy(&ds->ds_prop_cbs);
573			list_destroy(&ds->ds_sendstreams);
574			mutex_destroy(&ds->ds_lock);
575			mutex_destroy(&ds->ds_opening_lock);
576			mutex_destroy(&ds->ds_sendstream_lock);
577			mutex_destroy(&ds->ds_remap_deadlist_lock);
578			zfs_refcount_destroy(&ds->ds_longholds);
579			rrw_destroy(&ds->ds_bp_rwlock);
580			kmem_free(ds, sizeof (dsl_dataset_t));
581			if (err != 0) {
582				dmu_buf_rele(dbuf, tag);
583				return (err);
584			}
585			ds = winner;
586		} else {
587			ds->ds_fsid_guid =
588			    unique_insert(dsl_dataset_phys(ds)->ds_fsid_guid);
589			if (ds->ds_fsid_guid !=
590			    dsl_dataset_phys(ds)->ds_fsid_guid) {
591				zfs_dbgmsg("ds_fsid_guid changed from "
592				    "%llx to %llx for pool %s dataset id %llu",
593				    (long long)
594				    dsl_dataset_phys(ds)->ds_fsid_guid,
595				    (long long)ds->ds_fsid_guid,
596				    spa_name(dp->dp_spa),
597				    dsobj);
598			}
599		}
600	}
601
602	ASSERT3P(ds->ds_dbuf, ==, dbuf);
603	ASSERT3P(dsl_dataset_phys(ds), ==, dbuf->db_data);
604	ASSERT(dsl_dataset_phys(ds)->ds_prev_snap_obj != 0 ||
605	    spa_version(dp->dp_spa) < SPA_VERSION_ORIGIN ||
606	    dp->dp_origin_snap == NULL || ds == dp->dp_origin_snap);
607	*dsp = ds;
608
609	return (0);
610}
611
612int
613dsl_dataset_create_key_mapping(dsl_dataset_t *ds)
614{
615	dsl_dir_t *dd = ds->ds_dir;
616
617	if (dd->dd_crypto_obj == 0)
618		return (0);
619
620	return (spa_keystore_create_mapping(dd->dd_pool->dp_spa,
621	    ds, ds, &ds->ds_key_mapping));
622}
623
624int
625dsl_dataset_hold_obj_flags(dsl_pool_t *dp, uint64_t dsobj,
626    ds_hold_flags_t flags, void *tag, dsl_dataset_t **dsp)
627{
628	int err;
629
630	err = dsl_dataset_hold_obj(dp, dsobj, tag, dsp);
631	if (err != 0)
632		return (err);
633
634	ASSERT3P(*dsp, !=, NULL);
635
636	if (flags & DS_HOLD_FLAG_DECRYPT) {
637		err = dsl_dataset_create_key_mapping(*dsp);
638		if (err != 0)
639			dsl_dataset_rele(*dsp, tag);
640	}
641
642	return (err);
643}
644
645int
646dsl_dataset_hold_flags(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
647    void *tag, dsl_dataset_t **dsp)
648{
649	dsl_dir_t *dd;
650	const char *snapname;
651	uint64_t obj;
652	int err = 0;
653	dsl_dataset_t *ds;
654
655	err = dsl_dir_hold(dp, name, FTAG, &dd, &snapname);
656	if (err != 0)
657		return (err);
658
659	ASSERT(dsl_pool_config_held(dp));
660	obj = dsl_dir_phys(dd)->dd_head_dataset_obj;
661	if (obj != 0)
662		err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag, &ds);
663	else
664		err = SET_ERROR(ENOENT);
665
666	/* we may be looking for a snapshot */
667	if (err == 0 && snapname != NULL) {
668		dsl_dataset_t *snap_ds;
669
670		if (*snapname++ != '@') {
671			dsl_dataset_rele_flags(ds, flags, tag);
672			dsl_dir_rele(dd, FTAG);
673			return (SET_ERROR(ENOENT));
674		}
675
676		dprintf("looking for snapshot '%s'\n", snapname);
677		err = dsl_dataset_snap_lookup(ds, snapname, &obj);
678		if (err == 0) {
679			err = dsl_dataset_hold_obj_flags(dp, obj, flags, tag,
680			    &snap_ds);
681		}
682		dsl_dataset_rele_flags(ds, flags, tag);
683
684		if (err == 0) {
685			mutex_enter(&snap_ds->ds_lock);
686			if (snap_ds->ds_snapname[0] == 0)
687				(void) strlcpy(snap_ds->ds_snapname, snapname,
688				    sizeof (snap_ds->ds_snapname));
689			mutex_exit(&snap_ds->ds_lock);
690			ds = snap_ds;
691		}
692	}
693	if (err == 0)
694		*dsp = ds;
695	dsl_dir_rele(dd, FTAG);
696	return (err);
697}
698
699int
700dsl_dataset_hold(dsl_pool_t *dp, const char *name, void *tag,
701    dsl_dataset_t **dsp)
702{
703	return (dsl_dataset_hold_flags(dp, name, 0, tag, dsp));
704}
705
706int
707dsl_dataset_own_obj(dsl_pool_t *dp, uint64_t dsobj, ds_hold_flags_t flags,
708    void *tag, dsl_dataset_t **dsp)
709{
710	int err = dsl_dataset_hold_obj_flags(dp, dsobj, flags, tag, dsp);
711	if (err != 0)
712		return (err);
713	if (!dsl_dataset_tryown(*dsp, tag)) {
714		dsl_dataset_rele_flags(*dsp, flags, tag);
715		*dsp = NULL;
716		return (SET_ERROR(EBUSY));
717	}
718	return (0);
719}
720
721int
722dsl_dataset_own(dsl_pool_t *dp, const char *name, ds_hold_flags_t flags,
723    void *tag, dsl_dataset_t **dsp)
724{
725	int err = dsl_dataset_hold_flags(dp, name, flags, tag, dsp);
726	if (err != 0)
727		return (err);
728	if (!dsl_dataset_tryown(*dsp, tag)) {
729		dsl_dataset_rele_flags(*dsp, flags, tag);
730		return (SET_ERROR(EBUSY));
731	}
732	return (0);
733}
734
735/*
736 * See the comment above dsl_pool_hold() for details.  In summary, a long
737 * hold is used to prevent destruction of a dataset while the pool hold
738 * is dropped, allowing other concurrent operations (e.g. spa_sync()).
739 *
740 * The dataset and pool must be held when this function is called.  After it
741 * is called, the pool hold may be released while the dataset is still held
742 * and accessed.
743 */
744void
745dsl_dataset_long_hold(dsl_dataset_t *ds, void *tag)
746{
747	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
748	(void) zfs_refcount_add(&ds->ds_longholds, tag);
749}
750
751void
752dsl_dataset_long_rele(dsl_dataset_t *ds, void *tag)
753{
754	(void) zfs_refcount_remove(&ds->ds_longholds, tag);
755}
756
757/* Return B_TRUE if there are any long holds on this dataset. */
758boolean_t
759dsl_dataset_long_held(dsl_dataset_t *ds)
760{
761	return (!zfs_refcount_is_zero(&ds->ds_longholds));
762}
763
764void
765dsl_dataset_name(dsl_dataset_t *ds, char *name)
766{
767	if (ds == NULL) {
768		(void) strcpy(name, "mos");
769	} else {
770		dsl_dir_name(ds->ds_dir, name);
771		VERIFY0(dsl_dataset_get_snapname(ds));
772		if (ds->ds_snapname[0]) {
773			VERIFY3U(strlcat(name, "@", ZFS_MAX_DATASET_NAME_LEN),
774			    <, ZFS_MAX_DATASET_NAME_LEN);
775			/*
776			 * We use a "recursive" mutex so that we
777			 * can call dprintf_ds() with ds_lock held.
778			 */
779			if (!MUTEX_HELD(&ds->ds_lock)) {
780				mutex_enter(&ds->ds_lock);
781				VERIFY3U(strlcat(name, ds->ds_snapname,
782				    ZFS_MAX_DATASET_NAME_LEN), <,
783				    ZFS_MAX_DATASET_NAME_LEN);
784				mutex_exit(&ds->ds_lock);
785			} else {
786				VERIFY3U(strlcat(name, ds->ds_snapname,
787				    ZFS_MAX_DATASET_NAME_LEN), <,
788				    ZFS_MAX_DATASET_NAME_LEN);
789			}
790		}
791	}
792}
793
794int
795dsl_dataset_namelen(dsl_dataset_t *ds)
796{
797	VERIFY0(dsl_dataset_get_snapname(ds));
798	mutex_enter(&ds->ds_lock);
799	int len = dsl_dir_namelen(ds->ds_dir) + 1 + strlen(ds->ds_snapname);
800	mutex_exit(&ds->ds_lock);
801	return (len);
802}
803
804void
805dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
806{
807	dmu_buf_rele(ds->ds_dbuf, tag);
808}
809
810void
811dsl_dataset_remove_key_mapping(dsl_dataset_t *ds)
812{
813	dsl_dir_t *dd = ds->ds_dir;
814
815	if (dd == NULL || dd->dd_crypto_obj == 0)
816		return;
817
818	(void) spa_keystore_remove_mapping(dd->dd_pool->dp_spa,
819	    ds->ds_object, ds);
820}
821
822void
823dsl_dataset_rele_flags(dsl_dataset_t *ds, ds_hold_flags_t flags, void *tag)
824{
825	if (flags & DS_HOLD_FLAG_DECRYPT)
826		dsl_dataset_remove_key_mapping(ds);
827
828	dsl_dataset_rele(ds, tag);
829}
830
831void
832dsl_dataset_disown(dsl_dataset_t *ds, ds_hold_flags_t flags, void *tag)
833{
834	ASSERT3P(ds->ds_owner, ==, tag);
835	ASSERT(ds->ds_dbuf != NULL);
836
837	mutex_enter(&ds->ds_lock);
838	ds->ds_owner = NULL;
839	mutex_exit(&ds->ds_lock);
840	dsl_dataset_long_rele(ds, tag);
841	dsl_dataset_rele_flags(ds, flags, tag);
842}
843
844boolean_t
845dsl_dataset_tryown(dsl_dataset_t *ds, void *tag)
846{
847	boolean_t gotit = FALSE;
848
849	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
850	mutex_enter(&ds->ds_lock);
851	if (ds->ds_owner == NULL && !DS_IS_INCONSISTENT(ds)) {
852		ds->ds_owner = tag;
853		dsl_dataset_long_hold(ds, tag);
854		gotit = TRUE;
855	}
856	mutex_exit(&ds->ds_lock);
857	return (gotit);
858}
859
860boolean_t
861dsl_dataset_has_owner(dsl_dataset_t *ds)
862{
863	boolean_t rv;
864	mutex_enter(&ds->ds_lock);
865	rv = (ds->ds_owner != NULL);
866	mutex_exit(&ds->ds_lock);
867	return (rv);
868}
869
870void
871dsl_dataset_activate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
872{
873	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
874	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
875	uint64_t zero = 0;
876
877	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
878
879	spa_feature_incr(spa, f, tx);
880	dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
881
882	VERIFY0(zap_add(mos, dsobj, spa_feature_table[f].fi_guid,
883	    sizeof (zero), 1, &zero, tx));
884}
885
886void
887dsl_dataset_deactivate_feature(uint64_t dsobj, spa_feature_t f, dmu_tx_t *tx)
888{
889	spa_t *spa = dmu_tx_pool(tx)->dp_spa;
890	objset_t *mos = dmu_tx_pool(tx)->dp_meta_objset;
891
892	VERIFY(spa_feature_table[f].fi_flags & ZFEATURE_FLAG_PER_DATASET);
893
894	VERIFY0(zap_remove(mos, dsobj, spa_feature_table[f].fi_guid, tx));
895	spa_feature_decr(spa, f, tx);
896}
897
898uint64_t
899dsl_dataset_create_sync_dd(dsl_dir_t *dd, dsl_dataset_t *origin,
900    dsl_crypto_params_t *dcp, uint64_t flags, dmu_tx_t *tx)
901{
902	dsl_pool_t *dp = dd->dd_pool;
903	dmu_buf_t *dbuf;
904	dsl_dataset_phys_t *dsphys;
905	uint64_t dsobj;
906	objset_t *mos = dp->dp_meta_objset;
907
908	if (origin == NULL)
909		origin = dp->dp_origin_snap;
910
911	ASSERT(origin == NULL || origin->ds_dir->dd_pool == dp);
912	ASSERT(origin == NULL || dsl_dataset_phys(origin)->ds_num_children > 0);
913	ASSERT(dmu_tx_is_syncing(tx));
914	ASSERT(dsl_dir_phys(dd)->dd_head_dataset_obj == 0);
915
916	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
917	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
918	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
919	dmu_buf_will_dirty(dbuf, tx);
920	dsphys = dbuf->db_data;
921	bzero(dsphys, sizeof (dsl_dataset_phys_t));
922	dsphys->ds_dir_obj = dd->dd_object;
923	dsphys->ds_flags = flags;
924	dsphys->ds_fsid_guid = unique_create();
925	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
926	    sizeof (dsphys->ds_guid));
927	dsphys->ds_snapnames_zapobj =
928	    zap_create_norm(mos, U8_TEXTPREP_TOUPPER, DMU_OT_DSL_DS_SNAP_MAP,
929	    DMU_OT_NONE, 0, tx);
930	dsphys->ds_creation_time = gethrestime_sec();
931	dsphys->ds_creation_txg = tx->tx_txg == TXG_INITIAL ? 1 : tx->tx_txg;
932
933	if (origin == NULL) {
934		dsphys->ds_deadlist_obj = dsl_deadlist_alloc(mos, tx);
935	} else {
936		dsl_dataset_t *ohds; /* head of the origin snapshot */
937
938		dsphys->ds_prev_snap_obj = origin->ds_object;
939		dsphys->ds_prev_snap_txg =
940		    dsl_dataset_phys(origin)->ds_creation_txg;
941		dsphys->ds_referenced_bytes =
942		    dsl_dataset_phys(origin)->ds_referenced_bytes;
943		dsphys->ds_compressed_bytes =
944		    dsl_dataset_phys(origin)->ds_compressed_bytes;
945		dsphys->ds_uncompressed_bytes =
946		    dsl_dataset_phys(origin)->ds_uncompressed_bytes;
947		rrw_enter(&origin->ds_bp_rwlock, RW_READER, FTAG);
948		dsphys->ds_bp = dsl_dataset_phys(origin)->ds_bp;
949		rrw_exit(&origin->ds_bp_rwlock, FTAG);
950
951		/*
952		 * Inherit flags that describe the dataset's contents
953		 * (INCONSISTENT) or properties (Case Insensitive).
954		 */
955		dsphys->ds_flags |= dsl_dataset_phys(origin)->ds_flags &
956		    (DS_FLAG_INCONSISTENT | DS_FLAG_CI_DATASET);
957
958		for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
959			if (origin->ds_feature_inuse[f])
960				dsl_dataset_activate_feature(dsobj, f, tx);
961		}
962
963		dmu_buf_will_dirty(origin->ds_dbuf, tx);
964		dsl_dataset_phys(origin)->ds_num_children++;
965
966		VERIFY0(dsl_dataset_hold_obj(dp,
967		    dsl_dir_phys(origin->ds_dir)->dd_head_dataset_obj,
968		    FTAG, &ohds));
969		dsphys->ds_deadlist_obj = dsl_deadlist_clone(&ohds->ds_deadlist,
970		    dsphys->ds_prev_snap_txg, dsphys->ds_prev_snap_obj, tx);
971		dsl_dataset_rele(ohds, FTAG);
972
973		if (spa_version(dp->dp_spa) >= SPA_VERSION_NEXT_CLONES) {
974			if (dsl_dataset_phys(origin)->ds_next_clones_obj == 0) {
975				dsl_dataset_phys(origin)->ds_next_clones_obj =
976				    zap_create(mos,
977				    DMU_OT_NEXT_CLONES, DMU_OT_NONE, 0, tx);
978			}
979			VERIFY0(zap_add_int(mos,
980			    dsl_dataset_phys(origin)->ds_next_clones_obj,
981			    dsobj, tx));
982		}
983
984		dmu_buf_will_dirty(dd->dd_dbuf, tx);
985		dsl_dir_phys(dd)->dd_origin_obj = origin->ds_object;
986		if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
987			if (dsl_dir_phys(origin->ds_dir)->dd_clones == 0) {
988				dmu_buf_will_dirty(origin->ds_dir->dd_dbuf, tx);
989				dsl_dir_phys(origin->ds_dir)->dd_clones =
990				    zap_create(mos,
991				    DMU_OT_DSL_CLONES, DMU_OT_NONE, 0, tx);
992			}
993			VERIFY0(zap_add_int(mos,
994			    dsl_dir_phys(origin->ds_dir)->dd_clones,
995			    dsobj, tx));
996		}
997	}
998
999	/* handle encryption */
1000	dsl_dataset_create_crypt_sync(dsobj, dd, origin, dcp, tx);
1001
1002	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1003		dsphys->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1004
1005	dmu_buf_rele(dbuf, FTAG);
1006
1007	dmu_buf_will_dirty(dd->dd_dbuf, tx);
1008	dsl_dir_phys(dd)->dd_head_dataset_obj = dsobj;
1009
1010	return (dsobj);
1011}
1012
1013static void
1014dsl_dataset_zero_zil(dsl_dataset_t *ds, dmu_tx_t *tx)
1015{
1016	objset_t *os;
1017
1018	VERIFY0(dmu_objset_from_ds(ds, &os));
1019	if (bcmp(&os->os_zil_header, &zero_zil, sizeof (zero_zil)) != 0) {
1020		dsl_pool_t *dp = ds->ds_dir->dd_pool;
1021		zio_t *zio;
1022
1023		bzero(&os->os_zil_header, sizeof (os->os_zil_header));
1024		if (os->os_encrypted)
1025			os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_TRUE;
1026
1027		zio = zio_root(dp->dp_spa, NULL, NULL, ZIO_FLAG_MUSTSUCCEED);
1028		dsl_dataset_sync(ds, zio, tx);
1029		VERIFY0(zio_wait(zio));
1030
1031		/* dsl_dataset_sync_done will drop this reference. */
1032		dmu_buf_add_ref(ds->ds_dbuf, ds);
1033		dsl_dataset_sync_done(ds, tx);
1034	}
1035}
1036
1037uint64_t
1038dsl_dataset_create_sync(dsl_dir_t *pdd, const char *lastname,
1039    dsl_dataset_t *origin, uint64_t flags, cred_t *cr,
1040    dsl_crypto_params_t *dcp, dmu_tx_t *tx)
1041{
1042	dsl_pool_t *dp = pdd->dd_pool;
1043	uint64_t dsobj, ddobj;
1044	dsl_dir_t *dd;
1045
1046	ASSERT(dmu_tx_is_syncing(tx));
1047	ASSERT(lastname[0] != '@');
1048
1049	ddobj = dsl_dir_create_sync(dp, pdd, lastname, tx);
1050	VERIFY0(dsl_dir_hold_obj(dp, ddobj, lastname, FTAG, &dd));
1051
1052	dsobj = dsl_dataset_create_sync_dd(dd, origin, dcp,
1053	    flags & ~DS_CREATE_FLAG_NODIRTY, tx);
1054
1055	dsl_deleg_set_create_perms(dd, tx, cr);
1056
1057	/*
1058	 * Since we're creating a new node we know it's a leaf, so we can
1059	 * initialize the counts if the limit feature is active.
1060	 */
1061	if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_FS_SS_LIMIT)) {
1062		uint64_t cnt = 0;
1063		objset_t *os = dd->dd_pool->dp_meta_objset;
1064
1065		dsl_dir_zapify(dd, tx);
1066		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_FILESYSTEM_COUNT,
1067		    sizeof (cnt), 1, &cnt, tx));
1068		VERIFY0(zap_add(os, dd->dd_object, DD_FIELD_SNAPSHOT_COUNT,
1069		    sizeof (cnt), 1, &cnt, tx));
1070	}
1071
1072	dsl_dir_rele(dd, FTAG);
1073
1074	/*
1075	 * If we are creating a clone, make sure we zero out any stale
1076	 * data from the origin snapshots zil header.
1077	 */
1078	if (origin != NULL && !(flags & DS_CREATE_FLAG_NODIRTY)) {
1079		dsl_dataset_t *ds;
1080
1081		VERIFY0(dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
1082		dsl_dataset_zero_zil(ds, tx);
1083		dsl_dataset_rele(ds, FTAG);
1084	}
1085
1086	return (dsobj);
1087}
1088
1089/*
1090 * The unique space in the head dataset can be calculated by subtracting
1091 * the space used in the most recent snapshot, that is still being used
1092 * in this file system, from the space currently in use.  To figure out
1093 * the space in the most recent snapshot still in use, we need to take
1094 * the total space used in the snapshot and subtract out the space that
1095 * has been freed up since the snapshot was taken.
1096 */
1097void
1098dsl_dataset_recalc_head_uniq(dsl_dataset_t *ds)
1099{
1100	uint64_t mrs_used;
1101	uint64_t dlused, dlcomp, dluncomp;
1102
1103	ASSERT(!ds->ds_is_snapshot);
1104
1105	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0)
1106		mrs_used = dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes;
1107	else
1108		mrs_used = 0;
1109
1110	dsl_deadlist_space(&ds->ds_deadlist, &dlused, &dlcomp, &dluncomp);
1111
1112	ASSERT3U(dlused, <=, mrs_used);
1113	dsl_dataset_phys(ds)->ds_unique_bytes =
1114	    dsl_dataset_phys(ds)->ds_referenced_bytes - (mrs_used - dlused);
1115
1116	if (spa_version(ds->ds_dir->dd_pool->dp_spa) >=
1117	    SPA_VERSION_UNIQUE_ACCURATE)
1118		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1119}
1120
1121void
1122dsl_dataset_remove_from_next_clones(dsl_dataset_t *ds, uint64_t obj,
1123    dmu_tx_t *tx)
1124{
1125	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1126	uint64_t count;
1127	int err;
1128
1129	ASSERT(dsl_dataset_phys(ds)->ds_num_children >= 2);
1130	err = zap_remove_int(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1131	    obj, tx);
1132	/*
1133	 * The err should not be ENOENT, but a bug in a previous version
1134	 * of the code could cause upgrade_clones_cb() to not set
1135	 * ds_next_snap_obj when it should, leading to a missing entry.
1136	 * If we knew that the pool was created after
1137	 * SPA_VERSION_NEXT_CLONES, we could assert that it isn't
1138	 * ENOENT.  However, at least we can check that we don't have
1139	 * too many entries in the next_clones_obj even after failing to
1140	 * remove this one.
1141	 */
1142	if (err != ENOENT)
1143		VERIFY0(err);
1144	ASSERT0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1145	    &count));
1146	ASSERT3U(count, <=, dsl_dataset_phys(ds)->ds_num_children - 2);
1147}
1148
1149
1150blkptr_t *
1151dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1152{
1153	return (&dsl_dataset_phys(ds)->ds_bp);
1154}
1155
1156spa_t *
1157dsl_dataset_get_spa(dsl_dataset_t *ds)
1158{
1159	return (ds->ds_dir->dd_pool->dp_spa);
1160}
1161
1162void
1163dsl_dataset_dirty(dsl_dataset_t *ds, dmu_tx_t *tx)
1164{
1165	dsl_pool_t *dp;
1166
1167	if (ds == NULL) /* this is the meta-objset */
1168		return;
1169
1170	ASSERT(ds->ds_objset != NULL);
1171
1172	if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0)
1173		panic("dirtying snapshot!");
1174
1175	/* Must not dirty a dataset in the same txg where it got snapshotted. */
1176	ASSERT3U(tx->tx_txg, >, dsl_dataset_phys(ds)->ds_prev_snap_txg);
1177
1178	dp = ds->ds_dir->dd_pool;
1179	if (txg_list_add(&dp->dp_dirty_datasets, ds, tx->tx_txg)) {
1180		objset_t *os = ds->ds_objset;
1181
1182		/* up the hold count until we can be written out */
1183		dmu_buf_add_ref(ds->ds_dbuf, ds);
1184
1185		/* if this dataset is encrypted, grab a reference to the DCK */
1186		if (ds->ds_dir->dd_crypto_obj != 0 &&
1187		    !os->os_raw_receive &&
1188		    !os->os_next_write_raw[tx->tx_txg & TXG_MASK]) {
1189			ASSERT3P(ds->ds_key_mapping, !=, NULL);
1190			key_mapping_add_ref(ds->ds_key_mapping, ds);
1191		}
1192	}
1193}
1194
1195boolean_t
1196dsl_dataset_is_dirty(dsl_dataset_t *ds)
1197{
1198	for (int t = 0; t < TXG_SIZE; t++) {
1199		if (txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1200		    ds, t))
1201			return (B_TRUE);
1202	}
1203	return (B_FALSE);
1204}
1205
1206static int
1207dsl_dataset_snapshot_reserve_space(dsl_dataset_t *ds, dmu_tx_t *tx)
1208{
1209	uint64_t asize;
1210
1211	if (!dmu_tx_is_syncing(tx))
1212		return (0);
1213
1214	/*
1215	 * If there's an fs-only reservation, any blocks that might become
1216	 * owned by the snapshot dataset must be accommodated by space
1217	 * outside of the reservation.
1218	 */
1219	ASSERT(ds->ds_reserved == 0 || DS_UNIQUE_IS_ACCURATE(ds));
1220	asize = MIN(dsl_dataset_phys(ds)->ds_unique_bytes, ds->ds_reserved);
1221	if (asize > dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE))
1222		return (SET_ERROR(ENOSPC));
1223
1224	/*
1225	 * Propagate any reserved space for this snapshot to other
1226	 * snapshot checks in this sync group.
1227	 */
1228	if (asize > 0)
1229		dsl_dir_willuse_space(ds->ds_dir, asize, tx);
1230
1231	return (0);
1232}
1233
1234int
1235dsl_dataset_snapshot_check_impl(dsl_dataset_t *ds, const char *snapname,
1236    dmu_tx_t *tx, boolean_t recv, uint64_t cnt, cred_t *cr)
1237{
1238	int error;
1239	uint64_t value;
1240
1241	ds->ds_trysnap_txg = tx->tx_txg;
1242
1243	if (!dmu_tx_is_syncing(tx))
1244		return (0);
1245
1246	/*
1247	 * We don't allow multiple snapshots of the same txg.  If there
1248	 * is already one, try again.
1249	 */
1250	if (dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg)
1251		return (SET_ERROR(EAGAIN));
1252
1253	/*
1254	 * Check for conflicting snapshot name.
1255	 */
1256	error = dsl_dataset_snap_lookup(ds, snapname, &value);
1257	if (error == 0)
1258		return (SET_ERROR(EEXIST));
1259	if (error != ENOENT)
1260		return (error);
1261
1262	/*
1263	 * We don't allow taking snapshots of inconsistent datasets, such as
1264	 * those into which we are currently receiving.  However, if we are
1265	 * creating this snapshot as part of a receive, this check will be
1266	 * executed atomically with respect to the completion of the receive
1267	 * itself but prior to the clearing of DS_FLAG_INCONSISTENT; in this
1268	 * case we ignore this, knowing it will be fixed up for us shortly in
1269	 * dmu_recv_end_sync().
1270	 */
1271	if (!recv && DS_IS_INCONSISTENT(ds))
1272		return (SET_ERROR(EBUSY));
1273
1274	/*
1275	 * Skip the check for temporary snapshots or if we have already checked
1276	 * the counts in dsl_dataset_snapshot_check. This means we really only
1277	 * check the count here when we're receiving a stream.
1278	 */
1279	if (cnt != 0 && cr != NULL) {
1280		error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1281		    ZFS_PROP_SNAPSHOT_LIMIT, NULL, cr);
1282		if (error != 0)
1283			return (error);
1284	}
1285
1286	error = dsl_dataset_snapshot_reserve_space(ds, tx);
1287	if (error != 0)
1288		return (error);
1289
1290	return (0);
1291}
1292
1293int
1294dsl_dataset_snapshot_check(void *arg, dmu_tx_t *tx)
1295{
1296	dsl_dataset_snapshot_arg_t *ddsa = arg;
1297	dsl_pool_t *dp = dmu_tx_pool(tx);
1298	nvpair_t *pair;
1299	int rv = 0;
1300
1301	/*
1302	 * Pre-compute how many total new snapshots will be created for each
1303	 * level in the tree and below. This is needed for validating the
1304	 * snapshot limit when either taking a recursive snapshot or when
1305	 * taking multiple snapshots.
1306	 *
1307	 * The problem is that the counts are not actually adjusted when
1308	 * we are checking, only when we finally sync. For a single snapshot,
1309	 * this is easy, the count will increase by 1 at each node up the tree,
1310	 * but its more complicated for the recursive/multiple snapshot case.
1311	 *
1312	 * The dsl_fs_ss_limit_check function does recursively check the count
1313	 * at each level up the tree but since it is validating each snapshot
1314	 * independently we need to be sure that we are validating the complete
1315	 * count for the entire set of snapshots. We do this by rolling up the
1316	 * counts for each component of the name into an nvlist and then
1317	 * checking each of those cases with the aggregated count.
1318	 *
1319	 * This approach properly handles not only the recursive snapshot
1320	 * case (where we get all of those on the ddsa_snaps list) but also
1321	 * the sibling case (e.g. snapshot a/b and a/c so that we will also
1322	 * validate the limit on 'a' using a count of 2).
1323	 *
1324	 * We validate the snapshot names in the third loop and only report
1325	 * name errors once.
1326	 */
1327	if (dmu_tx_is_syncing(tx)) {
1328		nvlist_t *cnt_track = NULL;
1329		cnt_track = fnvlist_alloc();
1330
1331		/* Rollup aggregated counts into the cnt_track list */
1332		for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1333		    pair != NULL;
1334		    pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1335			char *pdelim;
1336			uint64_t val;
1337			char nm[MAXPATHLEN];
1338
1339			(void) strlcpy(nm, nvpair_name(pair), sizeof (nm));
1340			pdelim = strchr(nm, '@');
1341			if (pdelim == NULL)
1342				continue;
1343			*pdelim = '\0';
1344
1345			do {
1346				if (nvlist_lookup_uint64(cnt_track, nm,
1347				    &val) == 0) {
1348					/* update existing entry */
1349					fnvlist_add_uint64(cnt_track, nm,
1350					    val + 1);
1351				} else {
1352					/* add to list */
1353					fnvlist_add_uint64(cnt_track, nm, 1);
1354				}
1355
1356				pdelim = strrchr(nm, '/');
1357				if (pdelim != NULL)
1358					*pdelim = '\0';
1359			} while (pdelim != NULL);
1360		}
1361
1362		/* Check aggregated counts at each level */
1363		for (pair = nvlist_next_nvpair(cnt_track, NULL);
1364		    pair != NULL; pair = nvlist_next_nvpair(cnt_track, pair)) {
1365			int error = 0;
1366			char *name;
1367			uint64_t cnt = 0;
1368			dsl_dataset_t *ds;
1369
1370			name = nvpair_name(pair);
1371			cnt = fnvpair_value_uint64(pair);
1372			ASSERT(cnt > 0);
1373
1374			error = dsl_dataset_hold(dp, name, FTAG, &ds);
1375			if (error == 0) {
1376				error = dsl_fs_ss_limit_check(ds->ds_dir, cnt,
1377				    ZFS_PROP_SNAPSHOT_LIMIT, NULL,
1378				    ddsa->ddsa_cr);
1379				dsl_dataset_rele(ds, FTAG);
1380			}
1381
1382			if (error != 0) {
1383				if (ddsa->ddsa_errors != NULL)
1384					fnvlist_add_int32(ddsa->ddsa_errors,
1385					    name, error);
1386				rv = error;
1387				/* only report one error for this check */
1388				break;
1389			}
1390		}
1391		nvlist_free(cnt_track);
1392	}
1393
1394	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1395	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1396		int error = 0;
1397		dsl_dataset_t *ds;
1398		char *name, *atp;
1399		char dsname[ZFS_MAX_DATASET_NAME_LEN];
1400
1401		name = nvpair_name(pair);
1402		if (strlen(name) >= ZFS_MAX_DATASET_NAME_LEN)
1403			error = SET_ERROR(ENAMETOOLONG);
1404		if (error == 0) {
1405			atp = strchr(name, '@');
1406			if (atp == NULL)
1407				error = SET_ERROR(EINVAL);
1408			if (error == 0)
1409				(void) strlcpy(dsname, name, atp - name + 1);
1410		}
1411		if (error == 0)
1412			error = dsl_dataset_hold(dp, dsname, FTAG, &ds);
1413		if (error == 0) {
1414			/* passing 0/NULL skips dsl_fs_ss_limit_check */
1415			error = dsl_dataset_snapshot_check_impl(ds,
1416			    atp + 1, tx, B_FALSE, 0, NULL);
1417			dsl_dataset_rele(ds, FTAG);
1418		}
1419
1420		if (error != 0) {
1421			if (ddsa->ddsa_errors != NULL) {
1422				fnvlist_add_int32(ddsa->ddsa_errors,
1423				    name, error);
1424			}
1425			rv = error;
1426		}
1427	}
1428
1429	return (rv);
1430}
1431
1432void
1433dsl_dataset_snapshot_sync_impl(dsl_dataset_t *ds, const char *snapname,
1434    dmu_tx_t *tx)
1435{
1436	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1437	dmu_buf_t *dbuf;
1438	dsl_dataset_phys_t *dsphys;
1439	uint64_t dsobj, crtxg;
1440	objset_t *mos = dp->dp_meta_objset;
1441	objset_t *os;
1442
1443	ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
1444
1445	/*
1446	 * If we are on an old pool, the zil must not be active, in which
1447	 * case it will be zeroed.  Usually zil_suspend() accomplishes this.
1448	 */
1449	ASSERT(spa_version(dmu_tx_pool(tx)->dp_spa) >= SPA_VERSION_FAST_SNAP ||
1450	    dmu_objset_from_ds(ds, &os) != 0 ||
1451	    bcmp(&os->os_phys->os_zil_header, &zero_zil,
1452	    sizeof (zero_zil)) == 0);
1453
1454	/* Should not snapshot a dirty dataset. */
1455	ASSERT(!txg_list_member(&ds->ds_dir->dd_pool->dp_dirty_datasets,
1456	    ds, tx->tx_txg));
1457
1458	dsl_fs_ss_count_adjust(ds->ds_dir, 1, DD_FIELD_SNAPSHOT_COUNT, tx);
1459
1460	/*
1461	 * The origin's ds_creation_txg has to be < TXG_INITIAL
1462	 */
1463	if (strcmp(snapname, ORIGIN_DIR_NAME) == 0)
1464		crtxg = 1;
1465	else
1466		crtxg = tx->tx_txg;
1467
1468	dsobj = dmu_object_alloc(mos, DMU_OT_DSL_DATASET, 0,
1469	    DMU_OT_DSL_DATASET, sizeof (dsl_dataset_phys_t), tx);
1470	VERIFY0(dmu_bonus_hold(mos, dsobj, FTAG, &dbuf));
1471	dmu_buf_will_dirty(dbuf, tx);
1472	dsphys = dbuf->db_data;
1473	bzero(dsphys, sizeof (dsl_dataset_phys_t));
1474	dsphys->ds_dir_obj = ds->ds_dir->dd_object;
1475	dsphys->ds_fsid_guid = unique_create();
1476	(void) random_get_pseudo_bytes((void*)&dsphys->ds_guid,
1477	    sizeof (dsphys->ds_guid));
1478	dsphys->ds_prev_snap_obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
1479	dsphys->ds_prev_snap_txg = dsl_dataset_phys(ds)->ds_prev_snap_txg;
1480	dsphys->ds_next_snap_obj = ds->ds_object;
1481	dsphys->ds_num_children = 1;
1482	dsphys->ds_creation_time = gethrestime_sec();
1483	dsphys->ds_creation_txg = crtxg;
1484	dsphys->ds_deadlist_obj = dsl_dataset_phys(ds)->ds_deadlist_obj;
1485	dsphys->ds_referenced_bytes = dsl_dataset_phys(ds)->ds_referenced_bytes;
1486	dsphys->ds_compressed_bytes = dsl_dataset_phys(ds)->ds_compressed_bytes;
1487	dsphys->ds_uncompressed_bytes =
1488	    dsl_dataset_phys(ds)->ds_uncompressed_bytes;
1489	dsphys->ds_flags = dsl_dataset_phys(ds)->ds_flags;
1490	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
1491	dsphys->ds_bp = dsl_dataset_phys(ds)->ds_bp;
1492	rrw_exit(&ds->ds_bp_rwlock, FTAG);
1493	dmu_buf_rele(dbuf, FTAG);
1494
1495	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1496		if (ds->ds_feature_inuse[f])
1497			dsl_dataset_activate_feature(dsobj, f, tx);
1498	}
1499
1500	ASSERT3U(ds->ds_prev != 0, ==,
1501	    dsl_dataset_phys(ds)->ds_prev_snap_obj != 0);
1502	if (ds->ds_prev) {
1503		uint64_t next_clones_obj =
1504		    dsl_dataset_phys(ds->ds_prev)->ds_next_clones_obj;
1505		ASSERT(dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1506		    ds->ds_object ||
1507		    dsl_dataset_phys(ds->ds_prev)->ds_num_children > 1);
1508		if (dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj ==
1509		    ds->ds_object) {
1510			dmu_buf_will_dirty(ds->ds_prev->ds_dbuf, tx);
1511			ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, ==,
1512			    dsl_dataset_phys(ds->ds_prev)->ds_creation_txg);
1513			dsl_dataset_phys(ds->ds_prev)->ds_next_snap_obj = dsobj;
1514		} else if (next_clones_obj != 0) {
1515			dsl_dataset_remove_from_next_clones(ds->ds_prev,
1516			    dsphys->ds_next_snap_obj, tx);
1517			VERIFY0(zap_add_int(mos,
1518			    next_clones_obj, dsobj, tx));
1519		}
1520	}
1521
1522	/*
1523	 * If we have a reference-reservation on this dataset, we will
1524	 * need to increase the amount of refreservation being charged
1525	 * since our unique space is going to zero.
1526	 */
1527	if (ds->ds_reserved) {
1528		int64_t delta;
1529		ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
1530		delta = MIN(dsl_dataset_phys(ds)->ds_unique_bytes,
1531		    ds->ds_reserved);
1532		dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV,
1533		    delta, 0, 0, tx);
1534	}
1535
1536	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1537	dsl_dataset_phys(ds)->ds_deadlist_obj =
1538	    dsl_deadlist_clone(&ds->ds_deadlist, UINT64_MAX,
1539	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
1540	dsl_deadlist_close(&ds->ds_deadlist);
1541	dsl_deadlist_open(&ds->ds_deadlist, mos,
1542	    dsl_dataset_phys(ds)->ds_deadlist_obj);
1543	dsl_deadlist_add_key(&ds->ds_deadlist,
1544	    dsl_dataset_phys(ds)->ds_prev_snap_txg, tx);
1545
1546	if (dsl_dataset_remap_deadlist_exists(ds)) {
1547		uint64_t remap_deadlist_obj =
1548		    dsl_dataset_get_remap_deadlist_object(ds);
1549		/*
1550		 * Move the remap_deadlist to the snapshot.  The head
1551		 * will create a new remap deadlist on demand, from
1552		 * dsl_dataset_block_remapped().
1553		 */
1554		dsl_dataset_unset_remap_deadlist_object(ds, tx);
1555		dsl_deadlist_close(&ds->ds_remap_deadlist);
1556
1557		dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
1558		VERIFY0(zap_add(mos, dsobj, DS_FIELD_REMAP_DEADLIST,
1559		    sizeof (remap_deadlist_obj), 1, &remap_deadlist_obj, tx));
1560	}
1561
1562	/*
1563	 * Create a ivset guid for this snapshot if the dataset is
1564	 * encrypted. This may be overridden by a raw receive. A
1565	 * previous implementation of this code did not have this
1566	 * field as part of the on-disk format for ZFS encryption
1567	 * (see errata #4). As part of the remediation for this
1568	 * issue, we ask the user to enable the bookmark_v2 feature
1569	 * which is now a dependency of the encryption feature. We
1570	 * use this as a heuristic to determine when the user has
1571	 * elected to correct any datasets created with the old code.
1572	 * As a result, we only do this step if the bookmark_v2
1573	 * feature is enabled, which limits the number of states a
1574	 * given pool / dataset can be in with regards to terms of
1575	 * correcting the issue.
1576	 */
1577	if (ds->ds_dir->dd_crypto_obj != 0 &&
1578	    spa_feature_is_enabled(dp->dp_spa, SPA_FEATURE_BOOKMARK_V2)) {
1579		uint64_t ivset_guid = unique_create();
1580
1581		dmu_object_zapify(mos, dsobj, DMU_OT_DSL_DATASET, tx);
1582		VERIFY0(zap_add(mos, dsobj, DS_FIELD_IVSET_GUID,
1583		    sizeof (ivset_guid), 1, &ivset_guid, tx));
1584	}
1585
1586	ASSERT3U(dsl_dataset_phys(ds)->ds_prev_snap_txg, <, tx->tx_txg);
1587	dsl_dataset_phys(ds)->ds_prev_snap_obj = dsobj;
1588	dsl_dataset_phys(ds)->ds_prev_snap_txg = crtxg;
1589	dsl_dataset_phys(ds)->ds_unique_bytes = 0;
1590
1591	if (spa_version(dp->dp_spa) >= SPA_VERSION_UNIQUE_ACCURATE)
1592		dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_UNIQUE_ACCURATE;
1593
1594	VERIFY0(zap_add(mos, dsl_dataset_phys(ds)->ds_snapnames_zapobj,
1595	    snapname, 8, 1, &dsobj, tx));
1596
1597	if (ds->ds_prev)
1598		dsl_dataset_rele(ds->ds_prev, ds);
1599	VERIFY0(dsl_dataset_hold_obj(dp,
1600	    dsl_dataset_phys(ds)->ds_prev_snap_obj, ds, &ds->ds_prev));
1601
1602	dsl_scan_ds_snapshotted(ds, tx);
1603
1604	dsl_dir_snap_cmtime_update(ds->ds_dir);
1605
1606	spa_history_log_internal_ds(ds->ds_prev, "snapshot", tx, "");
1607}
1608
1609void
1610dsl_dataset_snapshot_sync(void *arg, dmu_tx_t *tx)
1611{
1612	dsl_dataset_snapshot_arg_t *ddsa = arg;
1613	dsl_pool_t *dp = dmu_tx_pool(tx);
1614	nvpair_t *pair;
1615
1616	for (pair = nvlist_next_nvpair(ddsa->ddsa_snaps, NULL);
1617	    pair != NULL; pair = nvlist_next_nvpair(ddsa->ddsa_snaps, pair)) {
1618		dsl_dataset_t *ds;
1619		char *name, *atp;
1620		char dsname[ZFS_MAX_DATASET_NAME_LEN];
1621
1622		name = nvpair_name(pair);
1623		atp = strchr(name, '@');
1624		(void) strlcpy(dsname, name, atp - name + 1);
1625		VERIFY0(dsl_dataset_hold(dp, dsname, FTAG, &ds));
1626
1627		dsl_dataset_snapshot_sync_impl(ds, atp + 1, tx);
1628		if (ddsa->ddsa_props != NULL) {
1629			dsl_props_set_sync_impl(ds->ds_prev,
1630			    ZPROP_SRC_LOCAL, ddsa->ddsa_props, tx);
1631		}
1632		dsl_dataset_rele(ds, FTAG);
1633	}
1634}
1635
1636/*
1637 * The snapshots must all be in the same pool.
1638 * All-or-nothing: if there are any failures, nothing will be modified.
1639 */
1640int
1641dsl_dataset_snapshot(nvlist_t *snaps, nvlist_t *props, nvlist_t *errors)
1642{
1643	dsl_dataset_snapshot_arg_t ddsa;
1644	nvpair_t *pair;
1645	boolean_t needsuspend;
1646	int error;
1647	spa_t *spa;
1648	char *firstname;
1649	nvlist_t *suspended = NULL;
1650
1651	pair = nvlist_next_nvpair(snaps, NULL);
1652	if (pair == NULL)
1653		return (0);
1654	firstname = nvpair_name(pair);
1655
1656	error = spa_open(firstname, &spa, FTAG);
1657	if (error != 0)
1658		return (error);
1659	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1660	spa_close(spa, FTAG);
1661
1662	if (needsuspend) {
1663		suspended = fnvlist_alloc();
1664		for (pair = nvlist_next_nvpair(snaps, NULL); pair != NULL;
1665		    pair = nvlist_next_nvpair(snaps, pair)) {
1666			char fsname[ZFS_MAX_DATASET_NAME_LEN];
1667			char *snapname = nvpair_name(pair);
1668			char *atp;
1669			void *cookie;
1670
1671			atp = strchr(snapname, '@');
1672			if (atp == NULL) {
1673				error = SET_ERROR(EINVAL);
1674				break;
1675			}
1676			(void) strlcpy(fsname, snapname, atp - snapname + 1);
1677
1678			error = zil_suspend(fsname, &cookie);
1679			if (error != 0)
1680				break;
1681			fnvlist_add_uint64(suspended, fsname,
1682			    (uintptr_t)cookie);
1683		}
1684	}
1685
1686	ddsa.ddsa_snaps = snaps;
1687	ddsa.ddsa_props = props;
1688	ddsa.ddsa_errors = errors;
1689	ddsa.ddsa_cr = CRED();
1690
1691	if (error == 0) {
1692		error = dsl_sync_task(firstname, dsl_dataset_snapshot_check,
1693		    dsl_dataset_snapshot_sync, &ddsa,
1694		    fnvlist_num_pairs(snaps) * 3, ZFS_SPACE_CHECK_NORMAL);
1695	}
1696
1697	if (suspended != NULL) {
1698		for (pair = nvlist_next_nvpair(suspended, NULL); pair != NULL;
1699		    pair = nvlist_next_nvpair(suspended, pair)) {
1700			zil_resume((void *)(uintptr_t)
1701			    fnvpair_value_uint64(pair));
1702		}
1703		fnvlist_free(suspended);
1704	}
1705
1706	return (error);
1707}
1708
1709typedef struct dsl_dataset_snapshot_tmp_arg {
1710	const char *ddsta_fsname;
1711	const char *ddsta_snapname;
1712	minor_t ddsta_cleanup_minor;
1713	const char *ddsta_htag;
1714} dsl_dataset_snapshot_tmp_arg_t;
1715
1716static int
1717dsl_dataset_snapshot_tmp_check(void *arg, dmu_tx_t *tx)
1718{
1719	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1720	dsl_pool_t *dp = dmu_tx_pool(tx);
1721	dsl_dataset_t *ds;
1722	int error;
1723
1724	error = dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds);
1725	if (error != 0)
1726		return (error);
1727
1728	/* NULL cred means no limit check for tmp snapshot */
1729	error = dsl_dataset_snapshot_check_impl(ds, ddsta->ddsta_snapname,
1730	    tx, B_FALSE, 0, NULL);
1731	if (error != 0) {
1732		dsl_dataset_rele(ds, FTAG);
1733		return (error);
1734	}
1735
1736	if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS) {
1737		dsl_dataset_rele(ds, FTAG);
1738		return (SET_ERROR(ENOTSUP));
1739	}
1740	error = dsl_dataset_user_hold_check_one(NULL, ddsta->ddsta_htag,
1741	    B_TRUE, tx);
1742	if (error != 0) {
1743		dsl_dataset_rele(ds, FTAG);
1744		return (error);
1745	}
1746
1747	dsl_dataset_rele(ds, FTAG);
1748	return (0);
1749}
1750
1751static void
1752dsl_dataset_snapshot_tmp_sync(void *arg, dmu_tx_t *tx)
1753{
1754	dsl_dataset_snapshot_tmp_arg_t *ddsta = arg;
1755	dsl_pool_t *dp = dmu_tx_pool(tx);
1756	dsl_dataset_t *ds;
1757
1758	VERIFY0(dsl_dataset_hold(dp, ddsta->ddsta_fsname, FTAG, &ds));
1759
1760	dsl_dataset_snapshot_sync_impl(ds, ddsta->ddsta_snapname, tx);
1761	dsl_dataset_user_hold_sync_one(ds->ds_prev, ddsta->ddsta_htag,
1762	    ddsta->ddsta_cleanup_minor, gethrestime_sec(), tx);
1763	dsl_destroy_snapshot_sync_impl(ds->ds_prev, B_TRUE, tx);
1764
1765	dsl_dataset_rele(ds, FTAG);
1766}
1767
1768int
1769dsl_dataset_snapshot_tmp(const char *fsname, const char *snapname,
1770    minor_t cleanup_minor, const char *htag)
1771{
1772	dsl_dataset_snapshot_tmp_arg_t ddsta;
1773	int error;
1774	spa_t *spa;
1775	boolean_t needsuspend;
1776	void *cookie;
1777
1778	ddsta.ddsta_fsname = fsname;
1779	ddsta.ddsta_snapname = snapname;
1780	ddsta.ddsta_cleanup_minor = cleanup_minor;
1781	ddsta.ddsta_htag = htag;
1782
1783	error = spa_open(fsname, &spa, FTAG);
1784	if (error != 0)
1785		return (error);
1786	needsuspend = (spa_version(spa) < SPA_VERSION_FAST_SNAP);
1787	spa_close(spa, FTAG);
1788
1789	if (needsuspend) {
1790		error = zil_suspend(fsname, &cookie);
1791		if (error != 0)
1792			return (error);
1793	}
1794
1795	error = dsl_sync_task(fsname, dsl_dataset_snapshot_tmp_check,
1796	    dsl_dataset_snapshot_tmp_sync, &ddsta, 3, ZFS_SPACE_CHECK_RESERVED);
1797
1798	if (needsuspend)
1799		zil_resume(cookie);
1800	return (error);
1801}
1802
1803void
1804dsl_dataset_sync(dsl_dataset_t *ds, zio_t *zio, dmu_tx_t *tx)
1805{
1806	ASSERT(dmu_tx_is_syncing(tx));
1807	ASSERT(ds->ds_objset != NULL);
1808	ASSERT(dsl_dataset_phys(ds)->ds_next_snap_obj == 0);
1809
1810	/*
1811	 * in case we had to change ds_fsid_guid when we opened it,
1812	 * sync it out now.
1813	 */
1814	dmu_buf_will_dirty(ds->ds_dbuf, tx);
1815	dsl_dataset_phys(ds)->ds_fsid_guid = ds->ds_fsid_guid;
1816
1817	if (ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] != 0) {
1818		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1819		    ds->ds_object, DS_FIELD_RESUME_OBJECT, 8, 1,
1820		    &ds->ds_resume_object[tx->tx_txg & TXG_MASK], tx));
1821		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1822		    ds->ds_object, DS_FIELD_RESUME_OFFSET, 8, 1,
1823		    &ds->ds_resume_offset[tx->tx_txg & TXG_MASK], tx));
1824		VERIFY0(zap_update(tx->tx_pool->dp_meta_objset,
1825		    ds->ds_object, DS_FIELD_RESUME_BYTES, 8, 1,
1826		    &ds->ds_resume_bytes[tx->tx_txg & TXG_MASK], tx));
1827		ds->ds_resume_object[tx->tx_txg & TXG_MASK] = 0;
1828		ds->ds_resume_offset[tx->tx_txg & TXG_MASK] = 0;
1829		ds->ds_resume_bytes[tx->tx_txg & TXG_MASK] = 0;
1830	}
1831
1832	dmu_objset_sync(ds->ds_objset, zio, tx);
1833
1834	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
1835		if (ds->ds_feature_activation_needed[f]) {
1836			if (ds->ds_feature_inuse[f])
1837				continue;
1838			dsl_dataset_activate_feature(ds->ds_object, f, tx);
1839			ds->ds_feature_inuse[f] = B_TRUE;
1840		}
1841	}
1842}
1843
1844static int
1845deadlist_enqueue_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
1846{
1847	dsl_deadlist_t *dl = arg;
1848	dsl_deadlist_insert(dl, bp, tx);
1849	return (0);
1850}
1851
1852void
1853dsl_dataset_sync_done(dsl_dataset_t *ds, dmu_tx_t *tx)
1854{
1855	objset_t *os = ds->ds_objset;
1856
1857	bplist_iterate(&ds->ds_pending_deadlist,
1858	    deadlist_enqueue_cb, &ds->ds_deadlist, tx);
1859
1860	if (os->os_synced_dnodes != NULL) {
1861		multilist_destroy(os->os_synced_dnodes);
1862		os->os_synced_dnodes = NULL;
1863	}
1864
1865	if (os->os_encrypted)
1866		os->os_next_write_raw[tx->tx_txg & TXG_MASK] = B_FALSE;
1867	else
1868		ASSERT0(os->os_next_write_raw[tx->tx_txg & TXG_MASK]);
1869
1870	ASSERT(!dmu_objset_is_dirty(os, dmu_tx_get_txg(tx)));
1871
1872	dmu_buf_rele(ds->ds_dbuf, ds);
1873}
1874
1875int
1876get_clones_stat_impl(dsl_dataset_t *ds, nvlist_t *val)
1877{
1878	uint64_t count = 0;
1879	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
1880	zap_cursor_t zc;
1881	zap_attribute_t za;
1882
1883	ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool));
1884
1885	/*
1886	 * There may be missing entries in ds_next_clones_obj
1887	 * due to a bug in a previous version of the code.
1888	 * Only trust it if it has the right number of entries.
1889	 */
1890	if (dsl_dataset_phys(ds)->ds_next_clones_obj != 0) {
1891		VERIFY0(zap_count(mos, dsl_dataset_phys(ds)->ds_next_clones_obj,
1892		    &count));
1893	}
1894	if (count != dsl_dataset_phys(ds)->ds_num_children - 1) {
1895		return (ENOENT);
1896	}
1897	for (zap_cursor_init(&zc, mos,
1898	    dsl_dataset_phys(ds)->ds_next_clones_obj);
1899	    zap_cursor_retrieve(&zc, &za) == 0;
1900	    zap_cursor_advance(&zc)) {
1901		dsl_dataset_t *clone;
1902		char buf[ZFS_MAX_DATASET_NAME_LEN];
1903		VERIFY0(dsl_dataset_hold_obj(ds->ds_dir->dd_pool,
1904		    za.za_first_integer, FTAG, &clone));
1905		dsl_dir_name(clone->ds_dir, buf);
1906		fnvlist_add_boolean(val, buf);
1907		dsl_dataset_rele(clone, FTAG);
1908	}
1909	zap_cursor_fini(&zc);
1910	return (0);
1911}
1912
1913void
1914get_clones_stat(dsl_dataset_t *ds, nvlist_t *nv)
1915{
1916	nvlist_t *propval = fnvlist_alloc();
1917	nvlist_t *val;
1918
1919	/*
1920	 * We use nvlist_alloc() instead of fnvlist_alloc() because the
1921	 * latter would allocate the list with NV_UNIQUE_NAME flag.
1922	 * As a result, every time a clone name is appended to the list
1923	 * it would be (linearly) searched for for a duplicate name.
1924	 * We already know that all clone names must be unique and we
1925	 * want avoid the quadratic complexity of double-checking that
1926	 * because we can have a large number of clones.
1927	 */
1928	VERIFY0(nvlist_alloc(&val, 0, KM_SLEEP));
1929
1930	if (get_clones_stat_impl(ds, val) == 0) {
1931		fnvlist_add_nvlist(propval, ZPROP_VALUE, val);
1932		fnvlist_add_nvlist(nv, zfs_prop_to_name(ZFS_PROP_CLONES),
1933		    propval);
1934	}
1935
1936	nvlist_free(val);
1937	nvlist_free(propval);
1938}
1939
1940/*
1941 * Returns a string that represents the receive resume stats token. It should
1942 * be freed with strfree().
1943 */
1944char *
1945get_receive_resume_stats_impl(dsl_dataset_t *ds)
1946{
1947	dsl_pool_t *dp = ds->ds_dir->dd_pool;
1948
1949	if (dsl_dataset_has_resume_receive_state(ds)) {
1950		char *str;
1951		void *packed;
1952		uint8_t *compressed;
1953		uint64_t val;
1954		nvlist_t *token_nv = fnvlist_alloc();
1955		size_t packed_size, compressed_size;
1956
1957		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1958		    DS_FIELD_RESUME_FROMGUID, sizeof (val), 1, &val) == 0) {
1959			fnvlist_add_uint64(token_nv, "fromguid", val);
1960		}
1961		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1962		    DS_FIELD_RESUME_OBJECT, sizeof (val), 1, &val) == 0) {
1963			fnvlist_add_uint64(token_nv, "object", val);
1964		}
1965		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1966		    DS_FIELD_RESUME_OFFSET, sizeof (val), 1, &val) == 0) {
1967			fnvlist_add_uint64(token_nv, "offset", val);
1968		}
1969		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1970		    DS_FIELD_RESUME_BYTES, sizeof (val), 1, &val) == 0) {
1971			fnvlist_add_uint64(token_nv, "bytes", val);
1972		}
1973		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1974		    DS_FIELD_RESUME_TOGUID, sizeof (val), 1, &val) == 0) {
1975			fnvlist_add_uint64(token_nv, "toguid", val);
1976		}
1977		char buf[256];
1978		if (zap_lookup(dp->dp_meta_objset, ds->ds_object,
1979		    DS_FIELD_RESUME_TONAME, 1, sizeof (buf), buf) == 0) {
1980			fnvlist_add_string(token_nv, "toname", buf);
1981		}
1982		if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1983		    DS_FIELD_RESUME_LARGEBLOCK) == 0) {
1984			fnvlist_add_boolean(token_nv, "largeblockok");
1985		}
1986		if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1987		    DS_FIELD_RESUME_EMBEDOK) == 0) {
1988			fnvlist_add_boolean(token_nv, "embedok");
1989		}
1990		if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1991		    DS_FIELD_RESUME_COMPRESSOK) == 0) {
1992			fnvlist_add_boolean(token_nv, "compressok");
1993		}
1994		if (zap_contains(dp->dp_meta_objset, ds->ds_object,
1995		    DS_FIELD_RESUME_RAWOK) == 0) {
1996			fnvlist_add_boolean(token_nv, "rawok");
1997		}
1998		packed = fnvlist_pack(token_nv, &packed_size);
1999		fnvlist_free(token_nv);
2000		compressed = kmem_alloc(packed_size, KM_SLEEP);
2001
2002		compressed_size = gzip_compress(packed, compressed,
2003		    packed_size, packed_size, 6);
2004
2005		zio_cksum_t cksum;
2006		fletcher_4_native(compressed, compressed_size, NULL, &cksum);
2007
2008		str = kmem_alloc(compressed_size * 2 + 1, KM_SLEEP);
2009		for (int i = 0; i < compressed_size; i++) {
2010			(void) sprintf(str + i * 2, "%02x", compressed[i]);
2011		}
2012		str[compressed_size * 2] = '\0';
2013		char *propval = kmem_asprintf("%u-%llx-%llx-%s",
2014		    ZFS_SEND_RESUME_TOKEN_VERSION,
2015		    (longlong_t)cksum.zc_word[0],
2016		    (longlong_t)packed_size, str);
2017		kmem_free(packed, packed_size);
2018		kmem_free(str, compressed_size * 2 + 1);
2019		kmem_free(compressed, packed_size);
2020		return (propval);
2021	}
2022	return (strdup(""));
2023}
2024
2025/*
2026 * Returns a string that represents the receive resume stats token of the
2027 * dataset's child. It should be freed with strfree().
2028 */
2029char *
2030get_child_receive_stats(dsl_dataset_t *ds)
2031{
2032	char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
2033	dsl_dataset_t *recv_ds;
2034	dsl_dataset_name(ds, recvname);
2035	if (strlcat(recvname, "/", sizeof (recvname)) <
2036	    sizeof (recvname) &&
2037	    strlcat(recvname, recv_clone_name, sizeof (recvname)) <
2038	    sizeof (recvname) &&
2039	    dsl_dataset_hold(ds->ds_dir->dd_pool, recvname, FTAG,
2040	    &recv_ds)  == 0) {
2041		char *propval = get_receive_resume_stats_impl(recv_ds);
2042		dsl_dataset_rele(recv_ds, FTAG);
2043		return (propval);
2044	}
2045	return (strdup(""));
2046}
2047
2048static void
2049get_receive_resume_stats(dsl_dataset_t *ds, nvlist_t *nv)
2050{
2051	char *propval = get_receive_resume_stats_impl(ds);
2052	if (strcmp(propval, "") != 0) {
2053		dsl_prop_nvlist_add_string(nv,
2054		    ZFS_PROP_RECEIVE_RESUME_TOKEN, propval);
2055	} else {
2056		char *childval = get_child_receive_stats(ds);
2057		if (strcmp(childval, "") != 0) {
2058			dsl_prop_nvlist_add_string(nv,
2059			    ZFS_PROP_RECEIVE_RESUME_TOKEN, childval);
2060		}
2061		strfree(childval);
2062	}
2063	strfree(propval);
2064}
2065
2066uint64_t
2067dsl_get_refratio(dsl_dataset_t *ds)
2068{
2069	uint64_t ratio = dsl_dataset_phys(ds)->ds_compressed_bytes == 0 ? 100 :
2070	    (dsl_dataset_phys(ds)->ds_uncompressed_bytes * 100 /
2071	    dsl_dataset_phys(ds)->ds_compressed_bytes);
2072	return (ratio);
2073}
2074
2075uint64_t
2076dsl_get_logicalreferenced(dsl_dataset_t *ds)
2077{
2078	return (dsl_dataset_phys(ds)->ds_uncompressed_bytes);
2079}
2080
2081uint64_t
2082dsl_get_compressratio(dsl_dataset_t *ds)
2083{
2084	if (ds->ds_is_snapshot) {
2085		return (dsl_get_refratio(ds));
2086	} else {
2087		dsl_dir_t *dd = ds->ds_dir;
2088		mutex_enter(&dd->dd_lock);
2089		uint64_t val = dsl_dir_get_compressratio(dd);
2090		mutex_exit(&dd->dd_lock);
2091		return (val);
2092	}
2093}
2094
2095uint64_t
2096dsl_get_used(dsl_dataset_t *ds)
2097{
2098	if (ds->ds_is_snapshot) {
2099		return (dsl_dataset_phys(ds)->ds_unique_bytes);
2100	} else {
2101		dsl_dir_t *dd = ds->ds_dir;
2102		mutex_enter(&dd->dd_lock);
2103		uint64_t val = dsl_dir_get_used(dd);
2104		mutex_exit(&dd->dd_lock);
2105		return (val);
2106	}
2107}
2108
2109uint64_t
2110dsl_get_creation(dsl_dataset_t *ds)
2111{
2112	return (dsl_dataset_phys(ds)->ds_creation_time);
2113}
2114
2115uint64_t
2116dsl_get_creationtxg(dsl_dataset_t *ds)
2117{
2118	return (dsl_dataset_phys(ds)->ds_creation_txg);
2119}
2120
2121uint64_t
2122dsl_get_refquota(dsl_dataset_t *ds)
2123{
2124	return (ds->ds_quota);
2125}
2126
2127uint64_t
2128dsl_get_refreservation(dsl_dataset_t *ds)
2129{
2130	return (ds->ds_reserved);
2131}
2132
2133uint64_t
2134dsl_get_guid(dsl_dataset_t *ds)
2135{
2136	return (dsl_dataset_phys(ds)->ds_guid);
2137}
2138
2139uint64_t
2140dsl_get_unique(dsl_dataset_t *ds)
2141{
2142	return (dsl_dataset_phys(ds)->ds_unique_bytes);
2143}
2144
2145uint64_t
2146dsl_get_objsetid(dsl_dataset_t *ds)
2147{
2148	return (ds->ds_object);
2149}
2150
2151uint64_t
2152dsl_get_userrefs(dsl_dataset_t *ds)
2153{
2154	return (ds->ds_userrefs);
2155}
2156
2157uint64_t
2158dsl_get_defer_destroy(dsl_dataset_t *ds)
2159{
2160	return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2161}
2162
2163uint64_t
2164dsl_get_referenced(dsl_dataset_t *ds)
2165{
2166	return (dsl_dataset_phys(ds)->ds_referenced_bytes);
2167}
2168
2169uint64_t
2170dsl_get_numclones(dsl_dataset_t *ds)
2171{
2172	ASSERT(ds->ds_is_snapshot);
2173	return (dsl_dataset_phys(ds)->ds_num_children - 1);
2174}
2175
2176uint64_t
2177dsl_get_inconsistent(dsl_dataset_t *ds)
2178{
2179	return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ?
2180	    1 : 0);
2181}
2182
2183uint64_t
2184dsl_get_available(dsl_dataset_t *ds)
2185{
2186	uint64_t refdbytes = dsl_get_referenced(ds);
2187	uint64_t availbytes = dsl_dir_space_available(ds->ds_dir,
2188	    NULL, 0, TRUE);
2189	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
2190		availbytes +=
2191		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2192	}
2193	if (ds->ds_quota != 0) {
2194		/*
2195		 * Adjust available bytes according to refquota
2196		 */
2197		if (refdbytes < ds->ds_quota) {
2198			availbytes = MIN(availbytes,
2199			    ds->ds_quota - refdbytes);
2200		} else {
2201			availbytes = 0;
2202		}
2203	}
2204	return (availbytes);
2205}
2206
2207int
2208dsl_get_written(dsl_dataset_t *ds, uint64_t *written)
2209{
2210	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2211	dsl_dataset_t *prev;
2212	int err = dsl_dataset_hold_obj(dp,
2213	    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
2214	if (err == 0) {
2215		uint64_t comp, uncomp;
2216		err = dsl_dataset_space_written(prev, ds, written,
2217		    &comp, &uncomp);
2218		dsl_dataset_rele(prev, FTAG);
2219	}
2220	return (err);
2221}
2222
2223/*
2224 * 'snap' should be a buffer of size ZFS_MAX_DATASET_NAME_LEN.
2225 */
2226int
2227dsl_get_prev_snap(dsl_dataset_t *ds, char *snap)
2228{
2229	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2230	if (ds->ds_prev != NULL && ds->ds_prev != dp->dp_origin_snap) {
2231		dsl_dataset_name(ds->ds_prev, snap);
2232		return (0);
2233	} else {
2234		return (ENOENT);
2235	}
2236}
2237
2238/*
2239 * Returns the mountpoint property and source for the given dataset in the value
2240 * and source buffers. The value buffer must be at least as large as MAXPATHLEN
2241 * and the source buffer as least as large a ZFS_MAX_DATASET_NAME_LEN.
2242 * Returns 0 on success and an error on failure.
2243 */
2244int
2245dsl_get_mountpoint(dsl_dataset_t *ds, const char *dsname, char *value,
2246    char *source)
2247{
2248	int error;
2249	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2250
2251	/* Retrieve the mountpoint value stored in the zap opbject */
2252	error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_MOUNTPOINT), 1,
2253	    ZAP_MAXVALUELEN, value, source);
2254	if (error != 0) {
2255		return (error);
2256	}
2257
2258	/*
2259	 * Process the dsname and source to find the full mountpoint string.
2260	 * Can be skipped for 'legacy' or 'none'.
2261	 */
2262	if (value[0] == '/') {
2263		char *buf = kmem_alloc(ZAP_MAXVALUELEN, KM_SLEEP);
2264		char *root = buf;
2265		const char *relpath;
2266
2267		/*
2268		 * If we inherit the mountpoint, even from a dataset
2269		 * with a received value, the source will be the path of
2270		 * the dataset we inherit from. If source is
2271		 * ZPROP_SOURCE_VAL_RECVD, the received value is not
2272		 * inherited.
2273		 */
2274		if (strcmp(source, ZPROP_SOURCE_VAL_RECVD) == 0) {
2275			relpath = "";
2276		} else {
2277			ASSERT0(strncmp(dsname, source, strlen(source)));
2278			relpath = dsname + strlen(source);
2279			if (relpath[0] == '/')
2280				relpath++;
2281		}
2282
2283		spa_altroot(dp->dp_spa, root, ZAP_MAXVALUELEN);
2284
2285		/*
2286		 * Special case an alternate root of '/'. This will
2287		 * avoid having multiple leading slashes in the
2288		 * mountpoint path.
2289		 */
2290		if (strcmp(root, "/") == 0)
2291			root++;
2292
2293		/*
2294		 * If the mountpoint is '/' then skip over this
2295		 * if we are obtaining either an alternate root or
2296		 * an inherited mountpoint.
2297		 */
2298		char *mnt = value;
2299		if (value[1] == '\0' && (root[0] != '\0' ||
2300		    relpath[0] != '\0'))
2301			mnt = value + 1;
2302
2303		if (relpath[0] == '\0') {
2304			(void) snprintf(value, ZAP_MAXVALUELEN, "%s%s",
2305			    root, mnt);
2306		} else {
2307			(void) snprintf(value, ZAP_MAXVALUELEN, "%s%s%s%s",
2308			    root, mnt, relpath[0] == '@' ? "" : "/",
2309			    relpath);
2310		}
2311		kmem_free(buf, ZAP_MAXVALUELEN);
2312	}
2313
2314	return (0);
2315}
2316
2317void
2318dsl_dataset_stats(dsl_dataset_t *ds, nvlist_t *nv)
2319{
2320	int err;
2321	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2322
2323	ASSERT(dsl_pool_config_held(dp));
2324
2325	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRATIO,
2326	    dsl_get_refratio(ds));
2327	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_LOGICALREFERENCED,
2328	    dsl_get_logicalreferenced(ds));
2329	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_COMPRESSRATIO,
2330	    dsl_get_compressratio(ds));
2331	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USED,
2332	    dsl_get_used(ds));
2333
2334	if (ds->ds_is_snapshot) {
2335		get_clones_stat(ds, nv);
2336	} else {
2337		char buf[ZFS_MAX_DATASET_NAME_LEN];
2338		if (dsl_get_prev_snap(ds, buf) == 0)
2339			dsl_prop_nvlist_add_string(nv, ZFS_PROP_PREV_SNAP,
2340			    buf);
2341		dsl_dir_stats(ds->ds_dir, nv);
2342	}
2343
2344	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_AVAILABLE,
2345	    dsl_get_available(ds));
2346	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFERENCED,
2347	    dsl_get_referenced(ds));
2348	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATION,
2349	    dsl_get_creation(ds));
2350	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_CREATETXG,
2351	    dsl_get_creationtxg(ds));
2352	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFQUOTA,
2353	    dsl_get_refquota(ds));
2354	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_REFRESERVATION,
2355	    dsl_get_refreservation(ds));
2356	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_GUID,
2357	    dsl_get_guid(ds));
2358	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_UNIQUE,
2359	    dsl_get_unique(ds));
2360	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_OBJSETID,
2361	    dsl_get_objsetid(ds));
2362	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_USERREFS,
2363	    dsl_get_userrefs(ds));
2364	dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_DEFER_DESTROY,
2365	    DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2366	dsl_dataset_crypt_stats(ds, nv);
2367
2368	if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
2369		uint64_t written, comp, uncomp;
2370		dsl_pool_t *dp = ds->ds_dir->dd_pool;
2371		dsl_dataset_t *prev;
2372
2373		err = dsl_dataset_hold_obj(dp,
2374		    dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &prev);
2375		if (err == 0) {
2376			err = dsl_dataset_space_written(prev, ds, &written,
2377			    &comp, &uncomp);
2378			dsl_dataset_rele(prev, FTAG);
2379			if (err == 0) {
2380				dsl_prop_nvlist_add_uint64(nv, ZFS_PROP_WRITTEN,
2381				    written);
2382			}
2383		}
2384	}
2385
2386	if (!dsl_dataset_is_snapshot(ds)) {
2387		/*
2388		 * A failed "newfs" (e.g. full) resumable receive leaves
2389		 * the stats set on this dataset.  Check here for the prop.
2390		 */
2391		get_receive_resume_stats(ds, nv);
2392
2393		/*
2394		 * A failed incremental resumable receive leaves the
2395		 * stats set on our child named "%recv".  Check the child
2396		 * for the prop.
2397		 */
2398		/* 6 extra bytes for /%recv */
2399		char recvname[ZFS_MAX_DATASET_NAME_LEN + 6];
2400		dsl_dataset_t *recv_ds;
2401		dsl_dataset_name(ds, recvname);
2402		if (strlcat(recvname, "/", sizeof (recvname)) <
2403		    sizeof (recvname) &&
2404		    strlcat(recvname, recv_clone_name, sizeof (recvname)) <
2405		    sizeof (recvname) &&
2406		    dsl_dataset_hold(dp, recvname, FTAG, &recv_ds) == 0) {
2407			get_receive_resume_stats(recv_ds, nv);
2408			dsl_dataset_rele(recv_ds, FTAG);
2409		}
2410	}
2411}
2412
2413void
2414dsl_dataset_fast_stat(dsl_dataset_t *ds, dmu_objset_stats_t *stat)
2415{
2416	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2417	ASSERT(dsl_pool_config_held(dp));
2418
2419	stat->dds_creation_txg = dsl_get_creationtxg(ds);
2420	stat->dds_inconsistent = dsl_get_inconsistent(ds);
2421	stat->dds_guid = dsl_get_guid(ds);
2422	stat->dds_origin[0] = '\0';
2423	if (ds->ds_is_snapshot) {
2424		stat->dds_is_snapshot = B_TRUE;
2425		stat->dds_num_clones = dsl_get_numclones(ds);
2426	} else {
2427		stat->dds_is_snapshot = B_FALSE;
2428		stat->dds_num_clones = 0;
2429
2430		if (dsl_dir_is_clone(ds->ds_dir)) {
2431			dsl_dir_get_origin(ds->ds_dir, stat->dds_origin);
2432		}
2433	}
2434}
2435
2436uint64_t
2437dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2438{
2439	return (ds->ds_fsid_guid);
2440}
2441
2442void
2443dsl_dataset_space(dsl_dataset_t *ds,
2444    uint64_t *refdbytesp, uint64_t *availbytesp,
2445    uint64_t *usedobjsp, uint64_t *availobjsp)
2446{
2447	*refdbytesp = dsl_dataset_phys(ds)->ds_referenced_bytes;
2448	*availbytesp = dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE);
2449	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes)
2450		*availbytesp +=
2451		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes;
2452	if (ds->ds_quota != 0) {
2453		/*
2454		 * Adjust available bytes according to refquota
2455		 */
2456		if (*refdbytesp < ds->ds_quota)
2457			*availbytesp = MIN(*availbytesp,
2458			    ds->ds_quota - *refdbytesp);
2459		else
2460			*availbytesp = 0;
2461	}
2462	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2463	*usedobjsp = BP_GET_FILL(&dsl_dataset_phys(ds)->ds_bp);
2464	rrw_exit(&ds->ds_bp_rwlock, FTAG);
2465	*availobjsp = DN_MAX_OBJECT - *usedobjsp;
2466}
2467
2468boolean_t
2469dsl_dataset_modified_since_snap(dsl_dataset_t *ds, dsl_dataset_t *snap)
2470{
2471	dsl_pool_t *dp = ds->ds_dir->dd_pool;
2472	uint64_t birth;
2473
2474	ASSERT(dsl_pool_config_held(dp));
2475	if (snap == NULL)
2476		return (B_FALSE);
2477	rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
2478	birth = dsl_dataset_get_blkptr(ds)->blk_birth;
2479	rrw_exit(&ds->ds_bp_rwlock, FTAG);
2480	if (birth > dsl_dataset_phys(snap)->ds_creation_txg) {
2481		objset_t *os, *os_snap;
2482		/*
2483		 * It may be that only the ZIL differs, because it was
2484		 * reset in the head.  Don't count that as being
2485		 * modified.
2486		 */
2487		if (dmu_objset_from_ds(ds, &os) != 0)
2488			return (B_TRUE);
2489		if (dmu_objset_from_ds(snap, &os_snap) != 0)
2490			return (B_TRUE);
2491		return (bcmp(&os->os_phys->os_meta_dnode,
2492		    &os_snap->os_phys->os_meta_dnode,
2493		    sizeof (os->os_phys->os_meta_dnode)) != 0);
2494	}
2495	return (B_FALSE);
2496}
2497
2498typedef struct dsl_dataset_rename_snapshot_arg {
2499	const char *ddrsa_fsname;
2500	const char *ddrsa_oldsnapname;
2501	const char *ddrsa_newsnapname;
2502	boolean_t ddrsa_recursive;
2503	dmu_tx_t *ddrsa_tx;
2504} dsl_dataset_rename_snapshot_arg_t;
2505
2506/* ARGSUSED */
2507static int
2508dsl_dataset_rename_snapshot_check_impl(dsl_pool_t *dp,
2509    dsl_dataset_t *hds, void *arg)
2510{
2511	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2512	int error;
2513	uint64_t val;
2514
2515	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2516	if (error != 0) {
2517		/* ignore nonexistent snapshots */
2518		return (error == ENOENT ? 0 : error);
2519	}
2520
2521	/* new name should not exist */
2522	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_newsnapname, &val);
2523	if (error == 0)
2524		error = SET_ERROR(EEXIST);
2525	else if (error == ENOENT)
2526		error = 0;
2527
2528	/* dataset name + 1 for the "@" + the new snapshot name must fit */
2529	if (dsl_dir_namelen(hds->ds_dir) + 1 +
2530	    strlen(ddrsa->ddrsa_newsnapname) >= ZFS_MAX_DATASET_NAME_LEN)
2531		error = SET_ERROR(ENAMETOOLONG);
2532
2533	return (error);
2534}
2535
2536static int
2537dsl_dataset_rename_snapshot_check(void *arg, dmu_tx_t *tx)
2538{
2539	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2540	dsl_pool_t *dp = dmu_tx_pool(tx);
2541	dsl_dataset_t *hds;
2542	int error;
2543
2544	error = dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds);
2545	if (error != 0)
2546		return (error);
2547
2548	if (ddrsa->ddrsa_recursive) {
2549		error = dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2550		    dsl_dataset_rename_snapshot_check_impl, ddrsa,
2551		    DS_FIND_CHILDREN);
2552	} else {
2553		error = dsl_dataset_rename_snapshot_check_impl(dp, hds, ddrsa);
2554	}
2555	dsl_dataset_rele(hds, FTAG);
2556	return (error);
2557}
2558
2559static int
2560dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t *dp,
2561    dsl_dataset_t *hds, void *arg)
2562{
2563	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2564	dsl_dataset_t *ds;
2565	uint64_t val;
2566	dmu_tx_t *tx = ddrsa->ddrsa_tx;
2567	int error;
2568
2569	error = dsl_dataset_snap_lookup(hds, ddrsa->ddrsa_oldsnapname, &val);
2570	ASSERT(error == 0 || error == ENOENT);
2571	if (error == ENOENT) {
2572		/* ignore nonexistent snapshots */
2573		return (0);
2574	}
2575
2576	VERIFY0(dsl_dataset_hold_obj(dp, val, FTAG, &ds));
2577
2578	/* log before we change the name */
2579	spa_history_log_internal_ds(ds, "rename", tx,
2580	    "-> @%s", ddrsa->ddrsa_newsnapname);
2581
2582	VERIFY0(dsl_dataset_snap_remove(hds, ddrsa->ddrsa_oldsnapname, tx,
2583	    B_FALSE));
2584	mutex_enter(&ds->ds_lock);
2585	(void) strcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname);
2586	mutex_exit(&ds->ds_lock);
2587	VERIFY0(zap_add(dp->dp_meta_objset,
2588	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2589	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2590
2591	dsl_dataset_rele(ds, FTAG);
2592	return (0);
2593}
2594
2595static void
2596dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2597{
2598	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2599	dsl_pool_t *dp = dmu_tx_pool(tx);
2600	dsl_dataset_t *hds;
2601
2602	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2603	ddrsa->ddrsa_tx = tx;
2604	if (ddrsa->ddrsa_recursive) {
2605		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2606		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2607		    DS_FIND_CHILDREN));
2608	} else {
2609		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2610	}
2611	dsl_dataset_rele(hds, FTAG);
2612}
2613
2614int
2615dsl_dataset_rename_snapshot(const char *fsname,
2616    const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2617{
2618	dsl_dataset_rename_snapshot_arg_t ddrsa;
2619
2620	ddrsa.ddrsa_fsname = fsname;
2621	ddrsa.ddrsa_oldsnapname = oldsnapname;
2622	ddrsa.ddrsa_newsnapname = newsnapname;
2623	ddrsa.ddrsa_recursive = recursive;
2624
2625	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2626	    dsl_dataset_rename_snapshot_sync, &ddrsa,
2627	    1, ZFS_SPACE_CHECK_RESERVED));
2628}
2629
2630/*
2631 * If we're doing an ownership handoff, we need to make sure that there is
2632 * only one long hold on the dataset.  We're not allowed to change anything here
2633 * so we don't permanently release the long hold or regular hold here.  We want
2634 * to do this only when syncing to avoid the dataset unexpectedly going away
2635 * when we release the long hold.
2636 */
2637static int
2638dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2639{
2640	boolean_t held;
2641
2642	if (!dmu_tx_is_syncing(tx))
2643		return (0);
2644
2645	if (owner != NULL) {
2646		VERIFY3P(ds->ds_owner, ==, owner);
2647		dsl_dataset_long_rele(ds, owner);
2648	}
2649
2650	held = dsl_dataset_long_held(ds);
2651
2652	if (owner != NULL)
2653		dsl_dataset_long_hold(ds, owner);
2654
2655	if (held)
2656		return (SET_ERROR(EBUSY));
2657
2658	return (0);
2659}
2660
2661int
2662dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2663{
2664	dsl_dataset_rollback_arg_t *ddra = arg;
2665	dsl_pool_t *dp = dmu_tx_pool(tx);
2666	dsl_dataset_t *ds;
2667	int64_t unused_refres_delta;
2668	int error;
2669
2670	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2671	if (error != 0)
2672		return (error);
2673
2674	/* must not be a snapshot */
2675	if (ds->ds_is_snapshot) {
2676		dsl_dataset_rele(ds, FTAG);
2677		return (SET_ERROR(EINVAL));
2678	}
2679
2680	/* must have a most recent snapshot */
2681	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2682		dsl_dataset_rele(ds, FTAG);
2683		return (SET_ERROR(ESRCH));
2684	}
2685
2686	/*
2687	 * No rollback to a snapshot created in the current txg, because
2688	 * the rollback may dirty the dataset and create blocks that are
2689	 * not reachable from the rootbp while having a birth txg that
2690	 * falls into the snapshot's range.
2691	 */
2692	if (dmu_tx_is_syncing(tx) &&
2693	    dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) {
2694		dsl_dataset_rele(ds, FTAG);
2695		return (SET_ERROR(EAGAIN));
2696	}
2697
2698	/*
2699	 * If the expected target snapshot is specified, then check that
2700	 * the latest snapshot is it.
2701	 */
2702	if (ddra->ddra_tosnap != NULL) {
2703		dsl_dataset_t *snapds;
2704
2705		/* Check if the target snapshot exists at all. */
2706		error = dsl_dataset_hold(dp, ddra->ddra_tosnap, FTAG, &snapds);
2707		if (error != 0) {
2708			/*
2709			 * ESRCH is used to signal that the target snapshot does
2710			 * not exist, while ENOENT is used to report that
2711			 * the rolled back dataset does not exist.
2712			 * ESRCH is also used to cover other cases where the
2713			 * target snapshot is not related to the dataset being
2714			 * rolled back such as being in a different pool.
2715			 */
2716			if (error == ENOENT || error == EXDEV)
2717				error = SET_ERROR(ESRCH);
2718			dsl_dataset_rele(ds, FTAG);
2719			return (error);
2720		}
2721		ASSERT(snapds->ds_is_snapshot);
2722
2723		/* Check if the snapshot is the latest snapshot indeed. */
2724		if (snapds != ds->ds_prev) {
2725			/*
2726			 * Distinguish between the case where the only problem
2727			 * is intervening snapshots (EEXIST) vs the snapshot
2728			 * not being a valid target for rollback (ESRCH).
2729			 */
2730			if (snapds->ds_dir == ds->ds_dir ||
2731			    (dsl_dir_is_clone(ds->ds_dir) &&
2732			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj ==
2733			    snapds->ds_object)) {
2734				error = SET_ERROR(EEXIST);
2735			} else {
2736				error = SET_ERROR(ESRCH);
2737			}
2738			dsl_dataset_rele(snapds, FTAG);
2739			dsl_dataset_rele(ds, FTAG);
2740			return (error);
2741		}
2742		dsl_dataset_rele(snapds, FTAG);
2743	}
2744
2745	/* must not have any bookmarks after the most recent snapshot */
2746	nvlist_t *proprequest = fnvlist_alloc();
2747	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2748	nvlist_t *bookmarks = fnvlist_alloc();
2749	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2750	fnvlist_free(proprequest);
2751	if (error != 0) {
2752		dsl_dataset_rele(ds, FTAG);
2753		return (error);
2754	}
2755	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2756	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2757		nvlist_t *valuenv =
2758		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2759		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
2760		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2761		if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2762			fnvlist_free(bookmarks);
2763			dsl_dataset_rele(ds, FTAG);
2764			return (SET_ERROR(EEXIST));
2765		}
2766	}
2767	fnvlist_free(bookmarks);
2768
2769	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2770	if (error != 0) {
2771		dsl_dataset_rele(ds, FTAG);
2772		return (error);
2773	}
2774
2775	/*
2776	 * Check if the snap we are rolling back to uses more than
2777	 * the refquota.
2778	 */
2779	if (ds->ds_quota != 0 &&
2780	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2781		dsl_dataset_rele(ds, FTAG);
2782		return (SET_ERROR(EDQUOT));
2783	}
2784
2785	/*
2786	 * When we do the clone swap, we will temporarily use more space
2787	 * due to the refreservation (the head will no longer have any
2788	 * unique space, so the entire amount of the refreservation will need
2789	 * to be free).  We will immediately destroy the clone, freeing
2790	 * this space, but the freeing happens over many txg's.
2791	 */
2792	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2793	    dsl_dataset_phys(ds)->ds_unique_bytes);
2794
2795	if (unused_refres_delta > 0 &&
2796	    unused_refres_delta >
2797	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2798		dsl_dataset_rele(ds, FTAG);
2799		return (SET_ERROR(ENOSPC));
2800	}
2801
2802	dsl_dataset_rele(ds, FTAG);
2803	return (0);
2804}
2805
2806void
2807dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2808{
2809	dsl_dataset_rollback_arg_t *ddra = arg;
2810	dsl_pool_t *dp = dmu_tx_pool(tx);
2811	dsl_dataset_t *ds, *clone;
2812	uint64_t cloneobj;
2813	char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2814
2815	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2816
2817	dsl_dataset_name(ds->ds_prev, namebuf);
2818	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2819
2820	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2821	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, NULL, tx);
2822
2823	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2824
2825	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2826	dsl_dataset_zero_zil(ds, tx);
2827
2828	dsl_destroy_head_sync_impl(clone, tx);
2829
2830	dsl_dataset_rele(clone, FTAG);
2831	dsl_dataset_rele(ds, FTAG);
2832}
2833
2834/*
2835 * Rolls back the given filesystem or volume to the most recent snapshot.
2836 * The name of the most recent snapshot will be returned under key "target"
2837 * in the result nvlist.
2838 *
2839 * If owner != NULL:
2840 * - The existing dataset MUST be owned by the specified owner at entry
2841 * - Upon return, dataset will still be held by the same owner, whether we
2842 *   succeed or not.
2843 *
2844 * This mode is required any time the existing filesystem is mounted.  See
2845 * notes above zfs_suspend_fs() for further details.
2846 */
2847int
2848dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner,
2849    nvlist_t *result)
2850{
2851	dsl_dataset_rollback_arg_t ddra;
2852
2853	ddra.ddra_fsname = fsname;
2854	ddra.ddra_tosnap = tosnap;
2855	ddra.ddra_owner = owner;
2856	ddra.ddra_result = result;
2857
2858	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2859	    dsl_dataset_rollback_sync, &ddra,
2860	    1, ZFS_SPACE_CHECK_RESERVED));
2861}
2862
2863struct promotenode {
2864	list_node_t link;
2865	dsl_dataset_t *ds;
2866};
2867
2868static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2869static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2870    void *tag);
2871static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2872
2873int
2874dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2875{
2876	dsl_dataset_promote_arg_t *ddpa = arg;
2877	dsl_pool_t *dp = dmu_tx_pool(tx);
2878	dsl_dataset_t *hds;
2879	struct promotenode *snap;
2880	dsl_dataset_t *origin_ds;
2881	int err;
2882	uint64_t unused;
2883	uint64_t ss_mv_cnt;
2884	size_t max_snap_len;
2885	boolean_t conflicting_snaps;
2886
2887	err = promote_hold(ddpa, dp, FTAG);
2888	if (err != 0)
2889		return (err);
2890
2891	hds = ddpa->ddpa_clone;
2892	snap = list_head(&ddpa->shared_snaps);
2893	origin_ds = snap->ds;
2894	max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2895
2896	snap = list_head(&ddpa->origin_snaps);
2897
2898	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2899		promote_rele(ddpa, FTAG);
2900		return (SET_ERROR(EXDEV));
2901	}
2902
2903	snap = list_head(&ddpa->shared_snaps);
2904	if (snap == NULL) {
2905		err = SET_ERROR(ENOENT);
2906		goto out;
2907	}
2908	origin_ds = snap->ds;
2909
2910	/*
2911	 * Encrypted clones share a DSL Crypto Key with their origin's dsl dir.
2912	 * When doing a promote we must make sure the encryption root for
2913	 * both the target and the target's origin does not change to avoid
2914	 * needing to rewrap encryption keys
2915	 */
2916	err = dsl_dataset_promote_crypt_check(hds->ds_dir, origin_ds->ds_dir);
2917	if (err != 0)
2918		goto out;
2919
2920	/*
2921	 * Compute and check the amount of space to transfer.  Since this is
2922	 * so expensive, don't do the preliminary check.
2923	 */
2924	if (!dmu_tx_is_syncing(tx)) {
2925		promote_rele(ddpa, FTAG);
2926		return (0);
2927	}
2928
2929	/* compute origin's new unique space */
2930	snap = list_tail(&ddpa->clone_snaps);
2931	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2932	    origin_ds->ds_object);
2933	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2934	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2935	    &ddpa->unique, &unused, &unused);
2936
2937	/*
2938	 * Walk the snapshots that we are moving
2939	 *
2940	 * Compute space to transfer.  Consider the incremental changes
2941	 * to used by each snapshot:
2942	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2943	 * So each snapshot gave birth to:
2944	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2945	 * So a sequence would look like:
2946	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2947	 * Which simplifies to:
2948	 * uN + kN + kN-1 + ... + k1 + k0
2949	 * Note however, if we stop before we reach the ORIGIN we get:
2950	 * uN + kN + kN-1 + ... + kM - uM-1
2951	 */
2952	conflicting_snaps = B_FALSE;
2953	ss_mv_cnt = 0;
2954	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2955	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2956	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2957	for (snap = list_head(&ddpa->shared_snaps); snap;
2958	    snap = list_next(&ddpa->shared_snaps, snap)) {
2959		uint64_t val, dlused, dlcomp, dluncomp;
2960		dsl_dataset_t *ds = snap->ds;
2961
2962		ss_mv_cnt++;
2963
2964		/*
2965		 * If there are long holds, we won't be able to evict
2966		 * the objset.
2967		 */
2968		if (dsl_dataset_long_held(ds)) {
2969			err = SET_ERROR(EBUSY);
2970			goto out;
2971		}
2972
2973		/* Check that the snapshot name does not conflict */
2974		VERIFY0(dsl_dataset_get_snapname(ds));
2975		if (strlen(ds->ds_snapname) >= max_snap_len) {
2976			err = SET_ERROR(ENAMETOOLONG);
2977			goto out;
2978		}
2979		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2980		if (err == 0) {
2981			fnvlist_add_boolean(ddpa->err_ds,
2982			    snap->ds->ds_snapname);
2983			conflicting_snaps = B_TRUE;
2984		} else if (err != ENOENT) {
2985			goto out;
2986		}
2987
2988		/* The very first snapshot does not have a deadlist */
2989		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2990			continue;
2991
2992		dsl_deadlist_space(&ds->ds_deadlist,
2993		    &dlused, &dlcomp, &dluncomp);
2994		ddpa->used += dlused;
2995		ddpa->comp += dlcomp;
2996		ddpa->uncomp += dluncomp;
2997	}
2998
2999	/*
3000	 * In order to return the full list of conflicting snapshots, we check
3001	 * whether there was a conflict after traversing all of them.
3002	 */
3003	if (conflicting_snaps) {
3004		err = SET_ERROR(EEXIST);
3005		goto out;
3006	}
3007
3008	/*
3009	 * If we are a clone of a clone then we never reached ORIGIN,
3010	 * so we need to subtract out the clone origin's used space.
3011	 */
3012	if (ddpa->origin_origin) {
3013		ddpa->used -=
3014		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
3015		ddpa->comp -=
3016		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
3017		ddpa->uncomp -=
3018		    dsl_dataset_phys(ddpa->origin_origin)->
3019		    ds_uncompressed_bytes;
3020	}
3021
3022	/* Check that there is enough space and limit headroom here */
3023	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
3024	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
3025	if (err != 0)
3026		goto out;
3027
3028	/*
3029	 * Compute the amounts of space that will be used by snapshots
3030	 * after the promotion (for both origin and clone).  For each,
3031	 * it is the amount of space that will be on all of their
3032	 * deadlists (that was not born before their new origin).
3033	 */
3034	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
3035		uint64_t space;
3036
3037		/*
3038		 * Note, typically this will not be a clone of a clone,
3039		 * so dd_origin_txg will be < TXG_INITIAL, so
3040		 * these snaplist_space() -> dsl_deadlist_space_range()
3041		 * calls will be fast because they do not have to
3042		 * iterate over all bps.
3043		 */
3044		snap = list_head(&ddpa->origin_snaps);
3045		err = snaplist_space(&ddpa->shared_snaps,
3046		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
3047		if (err != 0)
3048			goto out;
3049
3050		err = snaplist_space(&ddpa->clone_snaps,
3051		    snap->ds->ds_dir->dd_origin_txg, &space);
3052		if (err != 0)
3053			goto out;
3054		ddpa->cloneusedsnap += space;
3055	}
3056	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
3057	    DD_FLAG_USED_BREAKDOWN) {
3058		err = snaplist_space(&ddpa->origin_snaps,
3059		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
3060		    &ddpa->originusedsnap);
3061		if (err != 0)
3062			goto out;
3063	}
3064
3065out:
3066	promote_rele(ddpa, FTAG);
3067	return (err);
3068}
3069
3070void
3071dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
3072{
3073	dsl_dataset_promote_arg_t *ddpa = arg;
3074	dsl_pool_t *dp = dmu_tx_pool(tx);
3075	dsl_dataset_t *hds;
3076	struct promotenode *snap;
3077	dsl_dataset_t *origin_ds;
3078	dsl_dataset_t *origin_head;
3079	dsl_dir_t *dd;
3080	dsl_dir_t *odd = NULL;
3081	uint64_t oldnext_obj;
3082	int64_t delta;
3083
3084	VERIFY0(promote_hold(ddpa, dp, FTAG));
3085	hds = ddpa->ddpa_clone;
3086
3087	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
3088
3089	snap = list_head(&ddpa->shared_snaps);
3090	origin_ds = snap->ds;
3091	dd = hds->ds_dir;
3092
3093	snap = list_head(&ddpa->origin_snaps);
3094	origin_head = snap->ds;
3095
3096	/*
3097	 * We need to explicitly open odd, since origin_ds's dd will be
3098	 * changing.
3099	 */
3100	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
3101	    NULL, FTAG, &odd));
3102
3103	dsl_dataset_promote_crypt_sync(hds->ds_dir, odd, tx);
3104
3105	/* change origin's next snap */
3106	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
3107	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
3108	snap = list_tail(&ddpa->clone_snaps);
3109	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
3110	    origin_ds->ds_object);
3111	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
3112
3113	/* change the origin's next clone */
3114	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
3115		dsl_dataset_remove_from_next_clones(origin_ds,
3116		    snap->ds->ds_object, tx);
3117		VERIFY0(zap_add_int(dp->dp_meta_objset,
3118		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
3119		    oldnext_obj, tx));
3120	}
3121
3122	/* change origin */
3123	dmu_buf_will_dirty(dd->dd_dbuf, tx);
3124	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
3125	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
3126	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
3127	dmu_buf_will_dirty(odd->dd_dbuf, tx);
3128	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
3129	origin_head->ds_dir->dd_origin_txg =
3130	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
3131
3132	/* change dd_clone entries */
3133	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
3134		VERIFY0(zap_remove_int(dp->dp_meta_objset,
3135		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
3136		VERIFY0(zap_add_int(dp->dp_meta_objset,
3137		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
3138		    hds->ds_object, tx));
3139
3140		VERIFY0(zap_remove_int(dp->dp_meta_objset,
3141		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
3142		    origin_head->ds_object, tx));
3143		if (dsl_dir_phys(dd)->dd_clones == 0) {
3144			dsl_dir_phys(dd)->dd_clones =
3145			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
3146			    DMU_OT_NONE, 0, tx);
3147		}
3148		VERIFY0(zap_add_int(dp->dp_meta_objset,
3149		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
3150	}
3151
3152	/* move snapshots to this dir */
3153	for (snap = list_head(&ddpa->shared_snaps); snap;
3154	    snap = list_next(&ddpa->shared_snaps, snap)) {
3155		dsl_dataset_t *ds = snap->ds;
3156
3157		/*
3158		 * Property callbacks are registered to a particular
3159		 * dsl_dir.  Since ours is changing, evict the objset
3160		 * so that they will be unregistered from the old dsl_dir.
3161		 */
3162		if (ds->ds_objset) {
3163			dmu_objset_evict(ds->ds_objset);
3164			ds->ds_objset = NULL;
3165		}
3166
3167		/* move snap name entry */
3168		VERIFY0(dsl_dataset_get_snapname(ds));
3169		VERIFY0(dsl_dataset_snap_remove(origin_head,
3170		    ds->ds_snapname, tx, B_TRUE));
3171		VERIFY0(zap_add(dp->dp_meta_objset,
3172		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
3173		    8, 1, &ds->ds_object, tx));
3174		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
3175		    DD_FIELD_SNAPSHOT_COUNT, tx);
3176
3177		/* change containing dsl_dir */
3178		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3179		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
3180		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
3181		ASSERT3P(ds->ds_dir, ==, odd);
3182		dsl_dir_rele(ds->ds_dir, ds);
3183		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
3184		    NULL, ds, &ds->ds_dir));
3185
3186		/* move any clone references */
3187		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
3188		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
3189			zap_cursor_t zc;
3190			zap_attribute_t za;
3191
3192			for (zap_cursor_init(&zc, dp->dp_meta_objset,
3193			    dsl_dataset_phys(ds)->ds_next_clones_obj);
3194			    zap_cursor_retrieve(&zc, &za) == 0;
3195			    zap_cursor_advance(&zc)) {
3196				dsl_dataset_t *cnds;
3197				uint64_t o;
3198
3199				if (za.za_first_integer == oldnext_obj) {
3200					/*
3201					 * We've already moved the
3202					 * origin's reference.
3203					 */
3204					continue;
3205				}
3206
3207				VERIFY0(dsl_dataset_hold_obj(dp,
3208				    za.za_first_integer, FTAG, &cnds));
3209				o = dsl_dir_phys(cnds->ds_dir)->
3210				    dd_head_dataset_obj;
3211
3212				VERIFY0(zap_remove_int(dp->dp_meta_objset,
3213				    dsl_dir_phys(odd)->dd_clones, o, tx));
3214				VERIFY0(zap_add_int(dp->dp_meta_objset,
3215				    dsl_dir_phys(dd)->dd_clones, o, tx));
3216				dsl_dataset_rele(cnds, FTAG);
3217			}
3218			zap_cursor_fini(&zc);
3219		}
3220
3221		ASSERT(!dsl_prop_hascb(ds));
3222	}
3223
3224	/*
3225	 * Change space accounting.
3226	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
3227	 * both be valid, or both be 0 (resulting in delta == 0).  This
3228	 * is true for each of {clone,origin} independently.
3229	 */
3230
3231	delta = ddpa->cloneusedsnap -
3232	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
3233	ASSERT3S(delta, >=, 0);
3234	ASSERT3U(ddpa->used, >=, delta);
3235	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
3236	dsl_dir_diduse_space(dd, DD_USED_HEAD,
3237	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
3238
3239	delta = ddpa->originusedsnap -
3240	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
3241	ASSERT3S(delta, <=, 0);
3242	ASSERT3U(ddpa->used, >=, -delta);
3243	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
3244	dsl_dir_diduse_space(odd, DD_USED_HEAD,
3245	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
3246
3247	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
3248
3249	/* log history record */
3250	spa_history_log_internal_ds(hds, "promote", tx, "");
3251
3252	dsl_dir_rele(odd, FTAG);
3253	promote_rele(ddpa, FTAG);
3254}
3255
3256/*
3257 * Make a list of dsl_dataset_t's for the snapshots between first_obj
3258 * (exclusive) and last_obj (inclusive).  The list will be in reverse
3259 * order (last_obj will be the list_head()).  If first_obj == 0, do all
3260 * snapshots back to this dataset's origin.
3261 */
3262static int
3263snaplist_make(dsl_pool_t *dp,
3264    uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
3265{
3266	uint64_t obj = last_obj;
3267
3268	list_create(l, sizeof (struct promotenode),
3269	    offsetof(struct promotenode, link));
3270
3271	while (obj != first_obj) {
3272		dsl_dataset_t *ds;
3273		struct promotenode *snap;
3274		int err;
3275
3276		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
3277		ASSERT(err != ENOENT);
3278		if (err != 0)
3279			return (err);
3280
3281		if (first_obj == 0)
3282			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
3283
3284		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
3285		snap->ds = ds;
3286		list_insert_tail(l, snap);
3287		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3288	}
3289
3290	return (0);
3291}
3292
3293static int
3294snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
3295{
3296	struct promotenode *snap;
3297
3298	*spacep = 0;
3299	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
3300		uint64_t used, comp, uncomp;
3301		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3302		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
3303		*spacep += used;
3304	}
3305	return (0);
3306}
3307
3308static void
3309snaplist_destroy(list_t *l, void *tag)
3310{
3311	struct promotenode *snap;
3312
3313	if (l == NULL || !list_link_active(&l->list_head))
3314		return;
3315
3316	while ((snap = list_tail(l)) != NULL) {
3317		list_remove(l, snap);
3318		dsl_dataset_rele(snap->ds, tag);
3319		kmem_free(snap, sizeof (*snap));
3320	}
3321	list_destroy(l);
3322}
3323
3324static int
3325promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
3326{
3327	int error;
3328	dsl_dir_t *dd;
3329	struct promotenode *snap;
3330
3331	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
3332	    &ddpa->ddpa_clone);
3333	if (error != 0)
3334		return (error);
3335	dd = ddpa->ddpa_clone->ds_dir;
3336
3337	if (ddpa->ddpa_clone->ds_is_snapshot ||
3338	    !dsl_dir_is_clone(dd)) {
3339		dsl_dataset_rele(ddpa->ddpa_clone, tag);
3340		return (SET_ERROR(EINVAL));
3341	}
3342
3343	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
3344	    &ddpa->shared_snaps, tag);
3345	if (error != 0)
3346		goto out;
3347
3348	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
3349	    &ddpa->clone_snaps, tag);
3350	if (error != 0)
3351		goto out;
3352
3353	snap = list_head(&ddpa->shared_snaps);
3354	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
3355	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
3356	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
3357	    &ddpa->origin_snaps, tag);
3358	if (error != 0)
3359		goto out;
3360
3361	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
3362		error = dsl_dataset_hold_obj(dp,
3363		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
3364		    tag, &ddpa->origin_origin);
3365		if (error != 0)
3366			goto out;
3367	}
3368out:
3369	if (error != 0)
3370		promote_rele(ddpa, tag);
3371	return (error);
3372}
3373
3374static void
3375promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
3376{
3377	snaplist_destroy(&ddpa->shared_snaps, tag);
3378	snaplist_destroy(&ddpa->clone_snaps, tag);
3379	snaplist_destroy(&ddpa->origin_snaps, tag);
3380	if (ddpa->origin_origin != NULL)
3381		dsl_dataset_rele(ddpa->origin_origin, tag);
3382	dsl_dataset_rele(ddpa->ddpa_clone, tag);
3383}
3384
3385/*
3386 * Promote a clone.
3387 *
3388 * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
3389 * in with the name.  (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
3390 */
3391int
3392dsl_dataset_promote(const char *name, char *conflsnap)
3393{
3394	dsl_dataset_promote_arg_t ddpa = { 0 };
3395	uint64_t numsnaps;
3396	int error;
3397	nvpair_t *snap_pair;
3398	objset_t *os;
3399
3400	/*
3401	 * We will modify space proportional to the number of
3402	 * snapshots.  Compute numsnaps.
3403	 */
3404	error = dmu_objset_hold(name, FTAG, &os);
3405	if (error != 0)
3406		return (error);
3407	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
3408	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
3409	    &numsnaps);
3410	dmu_objset_rele(os, FTAG);
3411	if (error != 0)
3412		return (error);
3413
3414	ddpa.ddpa_clonename = name;
3415	ddpa.err_ds = fnvlist_alloc();
3416	ddpa.cr = CRED();
3417
3418	error = dsl_sync_task(name, dsl_dataset_promote_check,
3419	    dsl_dataset_promote_sync, &ddpa,
3420	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED);
3421
3422	/*
3423	 * Return the first conflicting snapshot found.
3424	 */
3425	snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL);
3426	if (snap_pair != NULL && conflsnap != NULL)
3427		(void) strcpy(conflsnap, nvpair_name(snap_pair));
3428
3429	fnvlist_free(ddpa.err_ds);
3430	return (error);
3431}
3432
3433int
3434dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
3435    dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
3436{
3437	/*
3438	 * "slack" factor for received datasets with refquota set on them.
3439	 * See the bottom of this function for details on its use.
3440	 */
3441	uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
3442	int64_t unused_refres_delta;
3443
3444	/* they should both be heads */
3445	if (clone->ds_is_snapshot ||
3446	    origin_head->ds_is_snapshot)
3447		return (SET_ERROR(EINVAL));
3448
3449	/* if we are not forcing, the branch point should be just before them */
3450	if (!force && clone->ds_prev != origin_head->ds_prev)
3451		return (SET_ERROR(EINVAL));
3452
3453	/* clone should be the clone (unless they are unrelated) */
3454	if (clone->ds_prev != NULL &&
3455	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
3456	    origin_head->ds_dir != clone->ds_prev->ds_dir)
3457		return (SET_ERROR(EINVAL));
3458
3459	/* the clone should be a child of the origin */
3460	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
3461		return (SET_ERROR(EINVAL));
3462
3463	/* origin_head shouldn't be modified unless 'force' */
3464	if (!force &&
3465	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
3466		return (SET_ERROR(ETXTBSY));
3467
3468	/* origin_head should have no long holds (e.g. is not mounted) */
3469	if (dsl_dataset_handoff_check(origin_head, owner, tx))
3470		return (SET_ERROR(EBUSY));
3471
3472	/* check amount of any unconsumed refreservation */
3473	unused_refres_delta =
3474	    (int64_t)MIN(origin_head->ds_reserved,
3475	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3476	    (int64_t)MIN(origin_head->ds_reserved,
3477	    dsl_dataset_phys(clone)->ds_unique_bytes);
3478
3479	if (unused_refres_delta > 0 &&
3480	    unused_refres_delta >
3481	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
3482		return (SET_ERROR(ENOSPC));
3483
3484	/*
3485	 * The clone can't be too much over the head's refquota.
3486	 *
3487	 * To ensure that the entire refquota can be used, we allow one
3488	 * transaction to exceed the the refquota.  Therefore, this check
3489	 * needs to also allow for the space referenced to be more than the
3490	 * refquota.  The maximum amount of space that one transaction can use
3491	 * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
3492	 * overage ensures that we are able to receive a filesystem that
3493	 * exceeds the refquota on the source system.
3494	 *
3495	 * So that overage is the refquota_slack we use below.
3496	 */
3497	if (origin_head->ds_quota != 0 &&
3498	    dsl_dataset_phys(clone)->ds_referenced_bytes >
3499	    origin_head->ds_quota + refquota_slack)
3500		return (SET_ERROR(EDQUOT));
3501
3502	return (0);
3503}
3504
3505static void
3506dsl_dataset_swap_remap_deadlists(dsl_dataset_t *clone,
3507    dsl_dataset_t *origin, dmu_tx_t *tx)
3508{
3509	uint64_t clone_remap_dl_obj, origin_remap_dl_obj;
3510	dsl_pool_t *dp = dmu_tx_pool(tx);
3511
3512	ASSERT(dsl_pool_sync_context(dp));
3513
3514	clone_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(clone);
3515	origin_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(origin);
3516
3517	if (clone_remap_dl_obj != 0) {
3518		dsl_deadlist_close(&clone->ds_remap_deadlist);
3519		dsl_dataset_unset_remap_deadlist_object(clone, tx);
3520	}
3521	if (origin_remap_dl_obj != 0) {
3522		dsl_deadlist_close(&origin->ds_remap_deadlist);
3523		dsl_dataset_unset_remap_deadlist_object(origin, tx);
3524	}
3525
3526	if (clone_remap_dl_obj != 0) {
3527		dsl_dataset_set_remap_deadlist_object(origin,
3528		    clone_remap_dl_obj, tx);
3529		dsl_deadlist_open(&origin->ds_remap_deadlist,
3530		    dp->dp_meta_objset, clone_remap_dl_obj);
3531	}
3532	if (origin_remap_dl_obj != 0) {
3533		dsl_dataset_set_remap_deadlist_object(clone,
3534		    origin_remap_dl_obj, tx);
3535		dsl_deadlist_open(&clone->ds_remap_deadlist,
3536		    dp->dp_meta_objset, origin_remap_dl_obj);
3537	}
3538}
3539
3540void
3541dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
3542    dsl_dataset_t *origin_head, dmu_tx_t *tx)
3543{
3544	dsl_pool_t *dp = dmu_tx_pool(tx);
3545	int64_t unused_refres_delta;
3546
3547	ASSERT(clone->ds_reserved == 0);
3548	/*
3549	 * NOTE: On DEBUG kernels there could be a race between this and
3550	 * the check function if spa_asize_inflation is adjusted...
3551	 */
3552	ASSERT(origin_head->ds_quota == 0 ||
3553	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
3554	    DMU_MAX_ACCESS * spa_asize_inflation);
3555	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
3556
3557	/*
3558	 * Swap per-dataset feature flags.
3559	 */
3560	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
3561		if (!(spa_feature_table[f].fi_flags &
3562		    ZFEATURE_FLAG_PER_DATASET)) {
3563			ASSERT(!clone->ds_feature_inuse[f]);
3564			ASSERT(!origin_head->ds_feature_inuse[f]);
3565			continue;
3566		}
3567
3568		boolean_t clone_inuse = clone->ds_feature_inuse[f];
3569		boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
3570
3571		if (clone_inuse) {
3572			dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
3573			clone->ds_feature_inuse[f] = B_FALSE;
3574		}
3575		if (origin_head_inuse) {
3576			dsl_dataset_deactivate_feature(origin_head->ds_object,
3577			    f, tx);
3578			origin_head->ds_feature_inuse[f] = B_FALSE;
3579		}
3580		if (clone_inuse) {
3581			dsl_dataset_activate_feature(origin_head->ds_object,
3582			    f, tx);
3583			origin_head->ds_feature_inuse[f] = B_TRUE;
3584		}
3585		if (origin_head_inuse) {
3586			dsl_dataset_activate_feature(clone->ds_object, f, tx);
3587			clone->ds_feature_inuse[f] = B_TRUE;
3588		}
3589	}
3590
3591	dmu_buf_will_dirty(clone->ds_dbuf, tx);
3592	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
3593
3594	if (clone->ds_objset != NULL) {
3595		dmu_objset_evict(clone->ds_objset);
3596		clone->ds_objset = NULL;
3597	}
3598
3599	if (origin_head->ds_objset != NULL) {
3600		dmu_objset_evict(origin_head->ds_objset);
3601		origin_head->ds_objset = NULL;
3602	}
3603
3604	unused_refres_delta =
3605	    (int64_t)MIN(origin_head->ds_reserved,
3606	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3607	    (int64_t)MIN(origin_head->ds_reserved,
3608	    dsl_dataset_phys(clone)->ds_unique_bytes);
3609
3610	/*
3611	 * Reset origin's unique bytes, if it exists.
3612	 */
3613	if (clone->ds_prev) {
3614		dsl_dataset_t *origin = clone->ds_prev;
3615		uint64_t comp, uncomp;
3616
3617		dmu_buf_will_dirty(origin->ds_dbuf, tx);
3618		dsl_deadlist_space_range(&clone->ds_deadlist,
3619		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
3620		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
3621	}
3622
3623	/* swap blkptrs */
3624	{
3625		rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG);
3626		rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG);
3627		blkptr_t tmp;
3628		tmp = dsl_dataset_phys(origin_head)->ds_bp;
3629		dsl_dataset_phys(origin_head)->ds_bp =
3630		    dsl_dataset_phys(clone)->ds_bp;
3631		dsl_dataset_phys(clone)->ds_bp = tmp;
3632		rrw_exit(&origin_head->ds_bp_rwlock, FTAG);
3633		rrw_exit(&clone->ds_bp_rwlock, FTAG);
3634	}
3635
3636	/* set dd_*_bytes */
3637	{
3638		int64_t dused, dcomp, duncomp;
3639		uint64_t cdl_used, cdl_comp, cdl_uncomp;
3640		uint64_t odl_used, odl_comp, odl_uncomp;
3641
3642		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
3643		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
3644
3645		dsl_deadlist_space(&clone->ds_deadlist,
3646		    &cdl_used, &cdl_comp, &cdl_uncomp);
3647		dsl_deadlist_space(&origin_head->ds_deadlist,
3648		    &odl_used, &odl_comp, &odl_uncomp);
3649
3650		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
3651		    cdl_used -
3652		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
3653		    odl_used);
3654		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
3655		    cdl_comp -
3656		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
3657		    odl_comp);
3658		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
3659		    cdl_uncomp -
3660		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
3661		    odl_uncomp);
3662
3663		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
3664		    dused, dcomp, duncomp, tx);
3665		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
3666		    -dused, -dcomp, -duncomp, tx);
3667
3668		/*
3669		 * The difference in the space used by snapshots is the
3670		 * difference in snapshot space due to the head's
3671		 * deadlist (since that's the only thing that's
3672		 * changing that affects the snapused).
3673		 */
3674		dsl_deadlist_space_range(&clone->ds_deadlist,
3675		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3676		    &cdl_used, &cdl_comp, &cdl_uncomp);
3677		dsl_deadlist_space_range(&origin_head->ds_deadlist,
3678		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3679		    &odl_used, &odl_comp, &odl_uncomp);
3680		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3681		    DD_USED_HEAD, DD_USED_SNAP, tx);
3682	}
3683
3684	/* swap ds_*_bytes */
3685	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3686	    dsl_dataset_phys(clone)->ds_referenced_bytes);
3687	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3688	    dsl_dataset_phys(clone)->ds_compressed_bytes);
3689	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3690	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3691	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3692	    dsl_dataset_phys(clone)->ds_unique_bytes);
3693
3694	/* apply any parent delta for change in unconsumed refreservation */
3695	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3696	    unused_refres_delta, 0, 0, tx);
3697
3698	/*
3699	 * Swap deadlists.
3700	 */
3701	dsl_deadlist_close(&clone->ds_deadlist);
3702	dsl_deadlist_close(&origin_head->ds_deadlist);
3703	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3704	    dsl_dataset_phys(clone)->ds_deadlist_obj);
3705	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3706	    dsl_dataset_phys(clone)->ds_deadlist_obj);
3707	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3708	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3709	dsl_dataset_swap_remap_deadlists(clone, origin_head, tx);
3710
3711	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3712
3713	spa_history_log_internal_ds(clone, "clone swap", tx,
3714	    "parent=%s", origin_head->ds_dir->dd_myname);
3715}
3716
3717/*
3718 * Given a pool name and a dataset object number in that pool,
3719 * return the name of that dataset.
3720 */
3721int
3722dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3723{
3724	dsl_pool_t *dp;
3725	dsl_dataset_t *ds;
3726	int error;
3727
3728	error = dsl_pool_hold(pname, FTAG, &dp);
3729	if (error != 0)
3730		return (error);
3731
3732	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3733	if (error == 0) {
3734		dsl_dataset_name(ds, buf);
3735		dsl_dataset_rele(ds, FTAG);
3736	}
3737	dsl_pool_rele(dp, FTAG);
3738
3739	return (error);
3740}
3741
3742int
3743dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3744    uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3745{
3746	int error = 0;
3747
3748	ASSERT3S(asize, >, 0);
3749
3750	/*
3751	 * *ref_rsrv is the portion of asize that will come from any
3752	 * unconsumed refreservation space.
3753	 */
3754	*ref_rsrv = 0;
3755
3756	mutex_enter(&ds->ds_lock);
3757	/*
3758	 * Make a space adjustment for reserved bytes.
3759	 */
3760	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3761		ASSERT3U(*used, >=,
3762		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3763		*used -=
3764		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3765		*ref_rsrv =
3766		    asize - MIN(asize, parent_delta(ds, asize + inflight));
3767	}
3768
3769	if (!check_quota || ds->ds_quota == 0) {
3770		mutex_exit(&ds->ds_lock);
3771		return (0);
3772	}
3773	/*
3774	 * If they are requesting more space, and our current estimate
3775	 * is over quota, they get to try again unless the actual
3776	 * on-disk is over quota and there are no pending changes (which
3777	 * may free up space for us).
3778	 */
3779	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3780	    ds->ds_quota) {
3781		if (inflight > 0 ||
3782		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3783			error = SET_ERROR(ERESTART);
3784		else
3785			error = SET_ERROR(EDQUOT);
3786	}
3787	mutex_exit(&ds->ds_lock);
3788
3789	return (error);
3790}
3791
3792typedef struct dsl_dataset_set_qr_arg {
3793	const char *ddsqra_name;
3794	zprop_source_t ddsqra_source;
3795	uint64_t ddsqra_value;
3796} dsl_dataset_set_qr_arg_t;
3797
3798
3799/* ARGSUSED */
3800static int
3801dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3802{
3803	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3804	dsl_pool_t *dp = dmu_tx_pool(tx);
3805	dsl_dataset_t *ds;
3806	int error;
3807	uint64_t newval;
3808
3809	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3810		return (SET_ERROR(ENOTSUP));
3811
3812	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3813	if (error != 0)
3814		return (error);
3815
3816	if (ds->ds_is_snapshot) {
3817		dsl_dataset_rele(ds, FTAG);
3818		return (SET_ERROR(EINVAL));
3819	}
3820
3821	error = dsl_prop_predict(ds->ds_dir,
3822	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3823	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3824	if (error != 0) {
3825		dsl_dataset_rele(ds, FTAG);
3826		return (error);
3827	}
3828
3829	if (newval == 0) {
3830		dsl_dataset_rele(ds, FTAG);
3831		return (0);
3832	}
3833
3834	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3835	    newval < ds->ds_reserved) {
3836		dsl_dataset_rele(ds, FTAG);
3837		return (SET_ERROR(ENOSPC));
3838	}
3839
3840	dsl_dataset_rele(ds, FTAG);
3841	return (0);
3842}
3843
3844static void
3845dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3846{
3847	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3848	dsl_pool_t *dp = dmu_tx_pool(tx);
3849	dsl_dataset_t *ds;
3850	uint64_t newval;
3851
3852	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3853
3854	dsl_prop_set_sync_impl(ds,
3855	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3856	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3857	    &ddsqra->ddsqra_value, tx);
3858
3859	VERIFY0(dsl_prop_get_int_ds(ds,
3860	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3861
3862	if (ds->ds_quota != newval) {
3863		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3864		ds->ds_quota = newval;
3865	}
3866	dsl_dataset_rele(ds, FTAG);
3867}
3868
3869int
3870dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3871    uint64_t refquota)
3872{
3873	dsl_dataset_set_qr_arg_t ddsqra;
3874
3875	ddsqra.ddsqra_name = dsname;
3876	ddsqra.ddsqra_source = source;
3877	ddsqra.ddsqra_value = refquota;
3878
3879	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3880	    dsl_dataset_set_refquota_sync, &ddsqra, 0,
3881	    ZFS_SPACE_CHECK_EXTRA_RESERVED));
3882}
3883
3884static int
3885dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3886{
3887	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3888	dsl_pool_t *dp = dmu_tx_pool(tx);
3889	dsl_dataset_t *ds;
3890	int error;
3891	uint64_t newval, unique;
3892
3893	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3894		return (SET_ERROR(ENOTSUP));
3895
3896	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3897	if (error != 0)
3898		return (error);
3899
3900	if (ds->ds_is_snapshot) {
3901		dsl_dataset_rele(ds, FTAG);
3902		return (SET_ERROR(EINVAL));
3903	}
3904
3905	error = dsl_prop_predict(ds->ds_dir,
3906	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3907	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3908	if (error != 0) {
3909		dsl_dataset_rele(ds, FTAG);
3910		return (error);
3911	}
3912
3913	/*
3914	 * If we are doing the preliminary check in open context, the
3915	 * space estimates may be inaccurate.
3916	 */
3917	if (!dmu_tx_is_syncing(tx)) {
3918		dsl_dataset_rele(ds, FTAG);
3919		return (0);
3920	}
3921
3922	mutex_enter(&ds->ds_lock);
3923	if (!DS_UNIQUE_IS_ACCURATE(ds))
3924		dsl_dataset_recalc_head_uniq(ds);
3925	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3926	mutex_exit(&ds->ds_lock);
3927
3928	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3929		uint64_t delta = MAX(unique, newval) -
3930		    MAX(unique, ds->ds_reserved);
3931
3932		if (delta >
3933		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3934		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3935			dsl_dataset_rele(ds, FTAG);
3936			return (SET_ERROR(ENOSPC));
3937		}
3938	}
3939
3940	dsl_dataset_rele(ds, FTAG);
3941	return (0);
3942}
3943
3944void
3945dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3946    zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3947{
3948	uint64_t newval;
3949	uint64_t unique;
3950	int64_t delta;
3951
3952	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3953	    source, sizeof (value), 1, &value, tx);
3954
3955	VERIFY0(dsl_prop_get_int_ds(ds,
3956	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3957
3958	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3959	mutex_enter(&ds->ds_dir->dd_lock);
3960	mutex_enter(&ds->ds_lock);
3961	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3962	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3963	delta = MAX(0, (int64_t)(newval - unique)) -
3964	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3965	ds->ds_reserved = newval;
3966	mutex_exit(&ds->ds_lock);
3967
3968	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3969	mutex_exit(&ds->ds_dir->dd_lock);
3970}
3971
3972static void
3973dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3974{
3975	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3976	dsl_pool_t *dp = dmu_tx_pool(tx);
3977	dsl_dataset_t *ds;
3978
3979	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3980	dsl_dataset_set_refreservation_sync_impl(ds,
3981	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3982	dsl_dataset_rele(ds, FTAG);
3983}
3984
3985int
3986dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3987    uint64_t refreservation)
3988{
3989	dsl_dataset_set_qr_arg_t ddsqra;
3990
3991	ddsqra.ddsqra_name = dsname;
3992	ddsqra.ddsqra_source = source;
3993	ddsqra.ddsqra_value = refreservation;
3994
3995	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3996	    dsl_dataset_set_refreservation_sync, &ddsqra, 0,
3997	    ZFS_SPACE_CHECK_EXTRA_RESERVED));
3998}
3999
4000/*
4001 * Return (in *usedp) the amount of space written in new that is not
4002 * present in oldsnap.  New may be a snapshot or the head.  Old must be
4003 * a snapshot before new, in new's filesystem (or its origin).  If not then
4004 * fail and return EINVAL.
4005 *
4006 * The written space is calculated by considering two components:  First, we
4007 * ignore any freed space, and calculate the written as new's used space
4008 * minus old's used space.  Next, we add in the amount of space that was freed
4009 * between the two snapshots, thus reducing new's used space relative to old's.
4010 * Specifically, this is the space that was born before old->ds_creation_txg,
4011 * and freed before new (ie. on new's deadlist or a previous deadlist).
4012 *
4013 * space freed                         [---------------------]
4014 * snapshots                       ---O-------O--------O-------O------
4015 *                                         oldsnap            new
4016 */
4017int
4018dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4019    uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4020{
4021	int err = 0;
4022	uint64_t snapobj;
4023	dsl_pool_t *dp = new->ds_dir->dd_pool;
4024
4025	ASSERT(dsl_pool_config_held(dp));
4026
4027	*usedp = 0;
4028	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
4029	*usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
4030
4031	*compp = 0;
4032	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
4033	*compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
4034
4035	*uncompp = 0;
4036	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
4037	*uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
4038
4039	snapobj = new->ds_object;
4040	while (snapobj != oldsnap->ds_object) {
4041		dsl_dataset_t *snap;
4042		uint64_t used, comp, uncomp;
4043
4044		if (snapobj == new->ds_object) {
4045			snap = new;
4046		} else {
4047			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4048			if (err != 0)
4049				break;
4050		}
4051
4052		if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
4053		    dsl_dataset_phys(oldsnap)->ds_creation_txg) {
4054			/*
4055			 * The blocks in the deadlist can not be born after
4056			 * ds_prev_snap_txg, so get the whole deadlist space,
4057			 * which is more efficient (especially for old-format
4058			 * deadlists).  Unfortunately the deadlist code
4059			 * doesn't have enough information to make this
4060			 * optimization itself.
4061			 */
4062			dsl_deadlist_space(&snap->ds_deadlist,
4063			    &used, &comp, &uncomp);
4064		} else {
4065			dsl_deadlist_space_range(&snap->ds_deadlist,
4066			    0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
4067			    &used, &comp, &uncomp);
4068		}
4069		*usedp += used;
4070		*compp += comp;
4071		*uncompp += uncomp;
4072
4073		/*
4074		 * If we get to the beginning of the chain of snapshots
4075		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4076		 * was not a snapshot of/before new.
4077		 */
4078		snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
4079		if (snap != new)
4080			dsl_dataset_rele(snap, FTAG);
4081		if (snapobj == 0) {
4082			err = SET_ERROR(EINVAL);
4083			break;
4084		}
4085
4086	}
4087	return (err);
4088}
4089
4090/*
4091 * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4092 * lastsnap, and all snapshots in between are deleted.
4093 *
4094 * blocks that would be freed            [---------------------------]
4095 * snapshots                       ---O-------O--------O-------O--------O
4096 *                                        firstsnap        lastsnap
4097 *
4098 * This is the set of blocks that were born after the snap before firstsnap,
4099 * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4100 * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4101 * We calculate this by iterating over the relevant deadlists (from the snap
4102 * after lastsnap, backward to the snap after firstsnap), summing up the
4103 * space on the deadlist that was born after the snap before firstsnap.
4104 */
4105int
4106dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4107    dsl_dataset_t *lastsnap,
4108    uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4109{
4110	int err = 0;
4111	uint64_t snapobj;
4112	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4113
4114	ASSERT(firstsnap->ds_is_snapshot);
4115	ASSERT(lastsnap->ds_is_snapshot);
4116
4117	/*
4118	 * Check that the snapshots are in the same dsl_dir, and firstsnap
4119	 * is before lastsnap.
4120	 */
4121	if (firstsnap->ds_dir != lastsnap->ds_dir ||
4122	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
4123	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
4124		return (SET_ERROR(EINVAL));
4125
4126	*usedp = *compp = *uncompp = 0;
4127
4128	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
4129	while (snapobj != firstsnap->ds_object) {
4130		dsl_dataset_t *ds;
4131		uint64_t used, comp, uncomp;
4132
4133		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4134		if (err != 0)
4135			break;
4136
4137		dsl_deadlist_space_range(&ds->ds_deadlist,
4138		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
4139		    &used, &comp, &uncomp);
4140		*usedp += used;
4141		*compp += comp;
4142		*uncompp += uncomp;
4143
4144		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
4145		ASSERT3U(snapobj, !=, 0);
4146		dsl_dataset_rele(ds, FTAG);
4147	}
4148	return (err);
4149}
4150
4151/*
4152 * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
4153 * For example, they could both be snapshots of the same filesystem, and
4154 * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
4155 * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
4156 * filesystem.  Or 'earlier' could be the origin's origin.
4157 *
4158 * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
4159 */
4160boolean_t
4161dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
4162    uint64_t earlier_txg)
4163{
4164	dsl_pool_t *dp = later->ds_dir->dd_pool;
4165	int error;
4166	boolean_t ret;
4167
4168	ASSERT(dsl_pool_config_held(dp));
4169	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
4170
4171	if (earlier_txg == 0)
4172		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
4173
4174	if (later->ds_is_snapshot &&
4175	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
4176		return (B_FALSE);
4177
4178	if (later->ds_dir == earlier->ds_dir)
4179		return (B_TRUE);
4180	if (!dsl_dir_is_clone(later->ds_dir))
4181		return (B_FALSE);
4182
4183	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
4184		return (B_TRUE);
4185	dsl_dataset_t *origin;
4186	error = dsl_dataset_hold_obj(dp,
4187	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
4188	if (error != 0)
4189		return (B_FALSE);
4190	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
4191	dsl_dataset_rele(origin, FTAG);
4192	return (ret);
4193}
4194
4195void
4196dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
4197{
4198	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
4199	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
4200}
4201
4202boolean_t
4203dsl_dataset_is_zapified(dsl_dataset_t *ds)
4204{
4205	dmu_object_info_t doi;
4206
4207	dmu_object_info_from_db(ds->ds_dbuf, &doi);
4208	return (doi.doi_type == DMU_OTN_ZAP_METADATA);
4209}
4210
4211boolean_t
4212dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
4213{
4214	return (dsl_dataset_is_zapified(ds) &&
4215	    zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
4216	    ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
4217}
4218
4219uint64_t
4220dsl_dataset_get_remap_deadlist_object(dsl_dataset_t *ds)
4221{
4222	uint64_t remap_deadlist_obj;
4223	int err;
4224
4225	if (!dsl_dataset_is_zapified(ds))
4226		return (0);
4227
4228	err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4229	    DS_FIELD_REMAP_DEADLIST, sizeof (remap_deadlist_obj), 1,
4230	    &remap_deadlist_obj);
4231
4232	if (err != 0) {
4233		VERIFY3S(err, ==, ENOENT);
4234		return (0);
4235	}
4236
4237	ASSERT(remap_deadlist_obj != 0);
4238	return (remap_deadlist_obj);
4239}
4240
4241boolean_t
4242dsl_dataset_remap_deadlist_exists(dsl_dataset_t *ds)
4243{
4244	EQUIV(dsl_deadlist_is_open(&ds->ds_remap_deadlist),
4245	    dsl_dataset_get_remap_deadlist_object(ds) != 0);
4246	return (dsl_deadlist_is_open(&ds->ds_remap_deadlist));
4247}
4248
4249static void
4250dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, uint64_t obj,
4251    dmu_tx_t *tx)
4252{
4253	ASSERT(obj != 0);
4254	dsl_dataset_zapify(ds, tx);
4255	VERIFY0(zap_add(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4256	    DS_FIELD_REMAP_DEADLIST, sizeof (obj), 1, &obj, tx));
4257}
4258
4259static void
4260dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, dmu_tx_t *tx)
4261{
4262	VERIFY0(zap_remove(ds->ds_dir->dd_pool->dp_meta_objset,
4263	    ds->ds_object, DS_FIELD_REMAP_DEADLIST, tx));
4264}
4265
4266void
4267dsl_dataset_destroy_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4268{
4269	uint64_t remap_deadlist_object;
4270	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4271
4272	ASSERT(dmu_tx_is_syncing(tx));
4273	ASSERT(dsl_dataset_remap_deadlist_exists(ds));
4274
4275	remap_deadlist_object = ds->ds_remap_deadlist.dl_object;
4276	dsl_deadlist_close(&ds->ds_remap_deadlist);
4277	dsl_deadlist_free(spa_meta_objset(spa), remap_deadlist_object, tx);
4278	dsl_dataset_unset_remap_deadlist_object(ds, tx);
4279	spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4280}
4281
4282void
4283dsl_dataset_create_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4284{
4285	uint64_t remap_deadlist_obj;
4286	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4287
4288	ASSERT(dmu_tx_is_syncing(tx));
4289	ASSERT(MUTEX_HELD(&ds->ds_remap_deadlist_lock));
4290	/*
4291	 * Currently we only create remap deadlists when there are indirect
4292	 * vdevs with referenced mappings.
4293	 */
4294	ASSERT(spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
4295
4296	remap_deadlist_obj = dsl_deadlist_clone(
4297	    &ds->ds_deadlist, UINT64_MAX,
4298	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
4299	dsl_dataset_set_remap_deadlist_object(ds,
4300	    remap_deadlist_obj, tx);
4301	dsl_deadlist_open(&ds->ds_remap_deadlist, spa_meta_objset(spa),
4302	    remap_deadlist_obj);
4303	spa_feature_incr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4304}
4305