xref: /illumos-gate/usr/src/uts/common/fs/zfs/dsl_dataset.c (revision 1bcef0f3e9dee4e6f3c617d8a92ac97c0ea47fb5)
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  */
73 int 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 
84 extern inline dsl_dataset_phys_t *dsl_dataset_phys(dsl_dataset_t *ds);
85 
86 static void dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds,
87     uint64_t obj, dmu_tx_t *tx);
88 static void dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds,
89     dmu_tx_t *tx);
90 
91 extern int spa_asize_inflation;
92 
93 static 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  */
100 static int64_t
parent_delta(dsl_dataset_t * ds,int64_t delta)101 parent_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 
117 void
dsl_dataset_block_born(dsl_dataset_t * ds,const blkptr_t * bp,dmu_tx_t * tx)118 dsl_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  */
171 void
dsl_dataset_block_remapped(dsl_dataset_t * ds,uint64_t vdev,uint64_t offset,uint64_t size,uint64_t birth,dmu_tx_t * tx)172 dsl_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 
205 int
dsl_dataset_block_kill(dsl_dataset_t * ds,const blkptr_t * bp,dmu_tx_t * tx,boolean_t async)206 dsl_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  */
297 static void
dsl_dataset_evict_sync(void * dbu)298 dsl_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 
307 static void
dsl_dataset_evict_async(void * dbu)308 dsl_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 
345 int
dsl_dataset_get_snapname(dsl_dataset_t * ds)346 dsl_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 
370 int
dsl_dataset_snap_lookup(dsl_dataset_t * ds,const char * name,uint64_t * value)371 dsl_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 
388 int
dsl_dataset_snap_remove(dsl_dataset_t * ds,const char * name,dmu_tx_t * tx,boolean_t adj_cnt)389 dsl_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 
413 boolean_t
dsl_dataset_try_add_ref(dsl_pool_t * dp,dsl_dataset_t * ds,void * tag)414 dsl_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 
431 int
dsl_dataset_hold_obj(dsl_pool_t * dp,uint64_t dsobj,void * tag,dsl_dataset_t ** dsp)432 dsl_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 
612 int
dsl_dataset_create_key_mapping(dsl_dataset_t * ds)613 dsl_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 
624 int
dsl_dataset_hold_obj_flags(dsl_pool_t * dp,uint64_t dsobj,ds_hold_flags_t flags,void * tag,dsl_dataset_t ** dsp)625 dsl_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 
645 int
dsl_dataset_hold_flags(dsl_pool_t * dp,const char * name,ds_hold_flags_t flags,void * tag,dsl_dataset_t ** dsp)646 dsl_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 
699 int
dsl_dataset_hold(dsl_pool_t * dp,const char * name,void * tag,dsl_dataset_t ** dsp)700 dsl_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 
706 int
dsl_dataset_own_obj(dsl_pool_t * dp,uint64_t dsobj,ds_hold_flags_t flags,void * tag,dsl_dataset_t ** dsp)707 dsl_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 
721 int
dsl_dataset_own(dsl_pool_t * dp,const char * name,ds_hold_flags_t flags,void * tag,dsl_dataset_t ** dsp)722 dsl_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  */
744 void
dsl_dataset_long_hold(dsl_dataset_t * ds,void * tag)745 dsl_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 
751 void
dsl_dataset_long_rele(dsl_dataset_t * ds,void * tag)752 dsl_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. */
758 boolean_t
dsl_dataset_long_held(dsl_dataset_t * ds)759 dsl_dataset_long_held(dsl_dataset_t *ds)
760 {
761 	return (!zfs_refcount_is_zero(&ds->ds_longholds));
762 }
763 
764 void
dsl_dataset_name(dsl_dataset_t * ds,char * name)765 dsl_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 
794 int
dsl_dataset_namelen(dsl_dataset_t * ds)795 dsl_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 
804 void
dsl_dataset_rele(dsl_dataset_t * ds,void * tag)805 dsl_dataset_rele(dsl_dataset_t *ds, void *tag)
806 {
807 	dmu_buf_rele(ds->ds_dbuf, tag);
808 }
809 
810 void
dsl_dataset_remove_key_mapping(dsl_dataset_t * ds)811 dsl_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 
822 void
dsl_dataset_rele_flags(dsl_dataset_t * ds,ds_hold_flags_t flags,void * tag)823 dsl_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 
831 void
dsl_dataset_disown(dsl_dataset_t * ds,ds_hold_flags_t flags,void * tag)832 dsl_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 
844 boolean_t
dsl_dataset_tryown(dsl_dataset_t * ds,void * tag)845 dsl_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 
860 boolean_t
dsl_dataset_has_owner(dsl_dataset_t * ds)861 dsl_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 
870 void
dsl_dataset_activate_feature(uint64_t dsobj,spa_feature_t f,dmu_tx_t * tx)871 dsl_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 
886 void
dsl_dataset_deactivate_feature(uint64_t dsobj,spa_feature_t f,dmu_tx_t * tx)887 dsl_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 
898 uint64_t
dsl_dataset_create_sync_dd(dsl_dir_t * dd,dsl_dataset_t * origin,dsl_crypto_params_t * dcp,uint64_t flags,dmu_tx_t * tx)899 dsl_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 
1013 static void
dsl_dataset_zero_zil(dsl_dataset_t * ds,dmu_tx_t * tx)1014 dsl_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 
1037 uint64_t
dsl_dataset_create_sync(dsl_dir_t * pdd,const char * lastname,dsl_dataset_t * origin,uint64_t flags,cred_t * cr,dsl_crypto_params_t * dcp,dmu_tx_t * tx)1038 dsl_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  */
1097 void
dsl_dataset_recalc_head_uniq(dsl_dataset_t * ds)1098 dsl_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 
1121 void
dsl_dataset_remove_from_next_clones(dsl_dataset_t * ds,uint64_t obj,dmu_tx_t * tx)1122 dsl_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 
1150 blkptr_t *
dsl_dataset_get_blkptr(dsl_dataset_t * ds)1151 dsl_dataset_get_blkptr(dsl_dataset_t *ds)
1152 {
1153 	return (&dsl_dataset_phys(ds)->ds_bp);
1154 }
1155 
1156 spa_t *
dsl_dataset_get_spa(dsl_dataset_t * ds)1157 dsl_dataset_get_spa(dsl_dataset_t *ds)
1158 {
1159 	return (ds->ds_dir->dd_pool->dp_spa);
1160 }
1161 
1162 void
dsl_dataset_dirty(dsl_dataset_t * ds,dmu_tx_t * tx)1163 dsl_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 
1195 boolean_t
dsl_dataset_is_dirty(dsl_dataset_t * ds)1196 dsl_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 
1206 static int
dsl_dataset_snapshot_reserve_space(dsl_dataset_t * ds,dmu_tx_t * tx)1207 dsl_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 
1234 int
dsl_dataset_snapshot_check_impl(dsl_dataset_t * ds,const char * snapname,dmu_tx_t * tx,boolean_t recv,uint64_t cnt,cred_t * cr)1235 dsl_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 
1293 int
dsl_dataset_snapshot_check(void * arg,dmu_tx_t * tx)1294 dsl_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 
1432 void
dsl_dataset_snapshot_sync_impl(dsl_dataset_t * ds,const char * snapname,dmu_tx_t * tx)1433 dsl_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 
1609 void
dsl_dataset_snapshot_sync(void * arg,dmu_tx_t * tx)1610 dsl_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  */
1640 int
dsl_dataset_snapshot(nvlist_t * snaps,nvlist_t * props,nvlist_t * errors)1641 dsl_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 
1709 typedef 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 
1716 static int
dsl_dataset_snapshot_tmp_check(void * arg,dmu_tx_t * tx)1717 dsl_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 
1751 static void
dsl_dataset_snapshot_tmp_sync(void * arg,dmu_tx_t * tx)1752 dsl_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 
1768 int
dsl_dataset_snapshot_tmp(const char * fsname,const char * snapname,minor_t cleanup_minor,const char * htag)1769 dsl_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 
1803 void
dsl_dataset_sync(dsl_dataset_t * ds,zio_t * zio,dmu_tx_t * tx)1804 dsl_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 
1844 static int
deadlist_enqueue_cb(void * arg,const blkptr_t * bp,dmu_tx_t * tx)1845 deadlist_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 
1852 void
dsl_dataset_sync_done(dsl_dataset_t * ds,dmu_tx_t * tx)1853 dsl_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 
1875 int
get_clones_stat_impl(dsl_dataset_t * ds,nvlist_t * val)1876 get_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 
1913 void
get_clones_stat(dsl_dataset_t * ds,nvlist_t * nv)1914 get_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  */
1944 char *
get_receive_resume_stats_impl(dsl_dataset_t * ds)1945 get_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_varsize(compressed, compressed_size, &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  */
2029 char *
get_child_receive_stats(dsl_dataset_t * ds)2030 get_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 
2048 static void
get_receive_resume_stats(dsl_dataset_t * ds,nvlist_t * nv)2049 get_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 
2066 uint64_t
dsl_get_refratio(dsl_dataset_t * ds)2067 dsl_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 
2075 uint64_t
dsl_get_logicalreferenced(dsl_dataset_t * ds)2076 dsl_get_logicalreferenced(dsl_dataset_t *ds)
2077 {
2078 	return (dsl_dataset_phys(ds)->ds_uncompressed_bytes);
2079 }
2080 
2081 uint64_t
dsl_get_compressratio(dsl_dataset_t * ds)2082 dsl_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 
2095 uint64_t
dsl_get_used(dsl_dataset_t * ds)2096 dsl_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 
2109 uint64_t
dsl_get_creation(dsl_dataset_t * ds)2110 dsl_get_creation(dsl_dataset_t *ds)
2111 {
2112 	return (dsl_dataset_phys(ds)->ds_creation_time);
2113 }
2114 
2115 uint64_t
dsl_get_creationtxg(dsl_dataset_t * ds)2116 dsl_get_creationtxg(dsl_dataset_t *ds)
2117 {
2118 	return (dsl_dataset_phys(ds)->ds_creation_txg);
2119 }
2120 
2121 uint64_t
dsl_get_refquota(dsl_dataset_t * ds)2122 dsl_get_refquota(dsl_dataset_t *ds)
2123 {
2124 	return (ds->ds_quota);
2125 }
2126 
2127 uint64_t
dsl_get_refreservation(dsl_dataset_t * ds)2128 dsl_get_refreservation(dsl_dataset_t *ds)
2129 {
2130 	return (ds->ds_reserved);
2131 }
2132 
2133 uint64_t
dsl_get_guid(dsl_dataset_t * ds)2134 dsl_get_guid(dsl_dataset_t *ds)
2135 {
2136 	return (dsl_dataset_phys(ds)->ds_guid);
2137 }
2138 
2139 uint64_t
dsl_get_unique(dsl_dataset_t * ds)2140 dsl_get_unique(dsl_dataset_t *ds)
2141 {
2142 	return (dsl_dataset_phys(ds)->ds_unique_bytes);
2143 }
2144 
2145 uint64_t
dsl_get_objsetid(dsl_dataset_t * ds)2146 dsl_get_objsetid(dsl_dataset_t *ds)
2147 {
2148 	return (ds->ds_object);
2149 }
2150 
2151 uint64_t
dsl_get_userrefs(dsl_dataset_t * ds)2152 dsl_get_userrefs(dsl_dataset_t *ds)
2153 {
2154 	return (ds->ds_userrefs);
2155 }
2156 
2157 uint64_t
dsl_get_defer_destroy(dsl_dataset_t * ds)2158 dsl_get_defer_destroy(dsl_dataset_t *ds)
2159 {
2160 	return (DS_IS_DEFER_DESTROY(ds) ? 1 : 0);
2161 }
2162 
2163 uint64_t
dsl_get_referenced(dsl_dataset_t * ds)2164 dsl_get_referenced(dsl_dataset_t *ds)
2165 {
2166 	return (dsl_dataset_phys(ds)->ds_referenced_bytes);
2167 }
2168 
2169 uint64_t
dsl_get_numclones(dsl_dataset_t * ds)2170 dsl_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 
2176 uint64_t
dsl_get_inconsistent(dsl_dataset_t * ds)2177 dsl_get_inconsistent(dsl_dataset_t *ds)
2178 {
2179 	return ((dsl_dataset_phys(ds)->ds_flags & DS_FLAG_INCONSISTENT) ?
2180 	    1 : 0);
2181 }
2182 
2183 uint64_t
dsl_get_available(dsl_dataset_t * ds)2184 dsl_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 
2207 int
dsl_get_written(dsl_dataset_t * ds,uint64_t * written)2208 dsl_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  */
2226 int
dsl_get_prev_snap(dsl_dataset_t * ds,char * snap)2227 dsl_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  */
2244 int
dsl_get_mountpoint(dsl_dataset_t * ds,const char * dsname,char * value,char * source)2245 dsl_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 
2317 void
dsl_dataset_stats(dsl_dataset_t * ds,nvlist_t * nv)2318 dsl_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 
2413 void
dsl_dataset_fast_stat(dsl_dataset_t * ds,dmu_objset_stats_t * stat)2414 dsl_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 
2436 uint64_t
dsl_dataset_fsid_guid(dsl_dataset_t * ds)2437 dsl_dataset_fsid_guid(dsl_dataset_t *ds)
2438 {
2439 	return (ds->ds_fsid_guid);
2440 }
2441 
2442 void
dsl_dataset_space(dsl_dataset_t * ds,uint64_t * refdbytesp,uint64_t * availbytesp,uint64_t * usedobjsp,uint64_t * availobjsp)2443 dsl_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 
2468 boolean_t
dsl_dataset_modified_since_snap(dsl_dataset_t * ds,dsl_dataset_t * snap)2469 dsl_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 
2498 typedef 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 */
2507 static int
dsl_dataset_rename_snapshot_check_impl(dsl_pool_t * dp,dsl_dataset_t * hds,void * arg)2508 dsl_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 
2536 static int
dsl_dataset_rename_snapshot_check(void * arg,dmu_tx_t * tx)2537 dsl_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 
2559 static int
dsl_dataset_rename_snapshot_sync_impl(dsl_pool_t * dp,dsl_dataset_t * hds,void * arg)2560 dsl_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) strlcpy(ds->ds_snapname, ddrsa->ddrsa_newsnapname,
2586 	    sizeof (ds->ds_snapname));
2587 	mutex_exit(&ds->ds_lock);
2588 	VERIFY0(zap_add(dp->dp_meta_objset,
2589 	    dsl_dataset_phys(hds)->ds_snapnames_zapobj,
2590 	    ds->ds_snapname, 8, 1, &ds->ds_object, tx));
2591 
2592 	dsl_dataset_rele(ds, FTAG);
2593 	return (0);
2594 }
2595 
2596 static void
dsl_dataset_rename_snapshot_sync(void * arg,dmu_tx_t * tx)2597 dsl_dataset_rename_snapshot_sync(void *arg, dmu_tx_t *tx)
2598 {
2599 	dsl_dataset_rename_snapshot_arg_t *ddrsa = arg;
2600 	dsl_pool_t *dp = dmu_tx_pool(tx);
2601 	dsl_dataset_t *hds;
2602 
2603 	VERIFY0(dsl_dataset_hold(dp, ddrsa->ddrsa_fsname, FTAG, &hds));
2604 	ddrsa->ddrsa_tx = tx;
2605 	if (ddrsa->ddrsa_recursive) {
2606 		VERIFY0(dmu_objset_find_dp(dp, hds->ds_dir->dd_object,
2607 		    dsl_dataset_rename_snapshot_sync_impl, ddrsa,
2608 		    DS_FIND_CHILDREN));
2609 	} else {
2610 		VERIFY0(dsl_dataset_rename_snapshot_sync_impl(dp, hds, ddrsa));
2611 	}
2612 	dsl_dataset_rele(hds, FTAG);
2613 }
2614 
2615 int
dsl_dataset_rename_snapshot(const char * fsname,const char * oldsnapname,const char * newsnapname,boolean_t recursive)2616 dsl_dataset_rename_snapshot(const char *fsname,
2617     const char *oldsnapname, const char *newsnapname, boolean_t recursive)
2618 {
2619 	dsl_dataset_rename_snapshot_arg_t ddrsa;
2620 
2621 	ddrsa.ddrsa_fsname = fsname;
2622 	ddrsa.ddrsa_oldsnapname = oldsnapname;
2623 	ddrsa.ddrsa_newsnapname = newsnapname;
2624 	ddrsa.ddrsa_recursive = recursive;
2625 
2626 	return (dsl_sync_task(fsname, dsl_dataset_rename_snapshot_check,
2627 	    dsl_dataset_rename_snapshot_sync, &ddrsa,
2628 	    1, ZFS_SPACE_CHECK_RESERVED));
2629 }
2630 
2631 /*
2632  * If we're doing an ownership handoff, we need to make sure that there is
2633  * only one long hold on the dataset.  We're not allowed to change anything here
2634  * so we don't permanently release the long hold or regular hold here.  We want
2635  * to do this only when syncing to avoid the dataset unexpectedly going away
2636  * when we release the long hold.
2637  */
2638 static int
dsl_dataset_handoff_check(dsl_dataset_t * ds,void * owner,dmu_tx_t * tx)2639 dsl_dataset_handoff_check(dsl_dataset_t *ds, void *owner, dmu_tx_t *tx)
2640 {
2641 	boolean_t held;
2642 
2643 	if (!dmu_tx_is_syncing(tx))
2644 		return (0);
2645 
2646 	if (owner != NULL) {
2647 		VERIFY3P(ds->ds_owner, ==, owner);
2648 		dsl_dataset_long_rele(ds, owner);
2649 	}
2650 
2651 	held = dsl_dataset_long_held(ds);
2652 
2653 	if (owner != NULL)
2654 		dsl_dataset_long_hold(ds, owner);
2655 
2656 	if (held)
2657 		return (SET_ERROR(EBUSY));
2658 
2659 	return (0);
2660 }
2661 
2662 int
dsl_dataset_rollback_check(void * arg,dmu_tx_t * tx)2663 dsl_dataset_rollback_check(void *arg, dmu_tx_t *tx)
2664 {
2665 	dsl_dataset_rollback_arg_t *ddra = arg;
2666 	dsl_pool_t *dp = dmu_tx_pool(tx);
2667 	dsl_dataset_t *ds;
2668 	int64_t unused_refres_delta;
2669 	int error;
2670 
2671 	error = dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds);
2672 	if (error != 0)
2673 		return (error);
2674 
2675 	/* must not be a snapshot */
2676 	if (ds->ds_is_snapshot) {
2677 		dsl_dataset_rele(ds, FTAG);
2678 		return (SET_ERROR(EINVAL));
2679 	}
2680 
2681 	/* must have a most recent snapshot */
2682 	if (dsl_dataset_phys(ds)->ds_prev_snap_txg < TXG_INITIAL) {
2683 		dsl_dataset_rele(ds, FTAG);
2684 		return (SET_ERROR(ESRCH));
2685 	}
2686 
2687 	/*
2688 	 * No rollback to a snapshot created in the current txg, because
2689 	 * the rollback may dirty the dataset and create blocks that are
2690 	 * not reachable from the rootbp while having a birth txg that
2691 	 * falls into the snapshot's range.
2692 	 */
2693 	if (dmu_tx_is_syncing(tx) &&
2694 	    dsl_dataset_phys(ds)->ds_prev_snap_txg >= tx->tx_txg) {
2695 		dsl_dataset_rele(ds, FTAG);
2696 		return (SET_ERROR(EAGAIN));
2697 	}
2698 
2699 	/*
2700 	 * If the expected target snapshot is specified, then check that
2701 	 * the latest snapshot is it.
2702 	 */
2703 	if (ddra->ddra_tosnap != NULL) {
2704 		dsl_dataset_t *snapds;
2705 
2706 		/* Check if the target snapshot exists at all. */
2707 		error = dsl_dataset_hold(dp, ddra->ddra_tosnap, FTAG, &snapds);
2708 		if (error != 0) {
2709 			/*
2710 			 * ESRCH is used to signal that the target snapshot does
2711 			 * not exist, while ENOENT is used to report that
2712 			 * the rolled back dataset does not exist.
2713 			 * ESRCH is also used to cover other cases where the
2714 			 * target snapshot is not related to the dataset being
2715 			 * rolled back such as being in a different pool.
2716 			 */
2717 			if (error == ENOENT || error == EXDEV)
2718 				error = SET_ERROR(ESRCH);
2719 			dsl_dataset_rele(ds, FTAG);
2720 			return (error);
2721 		}
2722 		ASSERT(snapds->ds_is_snapshot);
2723 
2724 		/* Check if the snapshot is the latest snapshot indeed. */
2725 		if (snapds != ds->ds_prev) {
2726 			/*
2727 			 * Distinguish between the case where the only problem
2728 			 * is intervening snapshots (EEXIST) vs the snapshot
2729 			 * not being a valid target for rollback (ESRCH).
2730 			 */
2731 			if (snapds->ds_dir == ds->ds_dir ||
2732 			    (dsl_dir_is_clone(ds->ds_dir) &&
2733 			    dsl_dir_phys(ds->ds_dir)->dd_origin_obj ==
2734 			    snapds->ds_object)) {
2735 				error = SET_ERROR(EEXIST);
2736 			} else {
2737 				error = SET_ERROR(ESRCH);
2738 			}
2739 			dsl_dataset_rele(snapds, FTAG);
2740 			dsl_dataset_rele(ds, FTAG);
2741 			return (error);
2742 		}
2743 		dsl_dataset_rele(snapds, FTAG);
2744 	}
2745 
2746 	/* must not have any bookmarks after the most recent snapshot */
2747 	nvlist_t *proprequest = fnvlist_alloc();
2748 	fnvlist_add_boolean(proprequest, zfs_prop_to_name(ZFS_PROP_CREATETXG));
2749 	nvlist_t *bookmarks = fnvlist_alloc();
2750 	error = dsl_get_bookmarks_impl(ds, proprequest, bookmarks);
2751 	fnvlist_free(proprequest);
2752 	if (error != 0) {
2753 		dsl_dataset_rele(ds, FTAG);
2754 		return (error);
2755 	}
2756 	for (nvpair_t *pair = nvlist_next_nvpair(bookmarks, NULL);
2757 	    pair != NULL; pair = nvlist_next_nvpair(bookmarks, pair)) {
2758 		nvlist_t *valuenv =
2759 		    fnvlist_lookup_nvlist(fnvpair_value_nvlist(pair),
2760 		    zfs_prop_to_name(ZFS_PROP_CREATETXG));
2761 		uint64_t createtxg = fnvlist_lookup_uint64(valuenv, "value");
2762 		if (createtxg > dsl_dataset_phys(ds)->ds_prev_snap_txg) {
2763 			fnvlist_free(bookmarks);
2764 			dsl_dataset_rele(ds, FTAG);
2765 			return (SET_ERROR(EEXIST));
2766 		}
2767 	}
2768 	fnvlist_free(bookmarks);
2769 
2770 	error = dsl_dataset_handoff_check(ds, ddra->ddra_owner, tx);
2771 	if (error != 0) {
2772 		dsl_dataset_rele(ds, FTAG);
2773 		return (error);
2774 	}
2775 
2776 	/*
2777 	 * Check if the snap we are rolling back to uses more than
2778 	 * the refquota.
2779 	 */
2780 	if (ds->ds_quota != 0 &&
2781 	    dsl_dataset_phys(ds->ds_prev)->ds_referenced_bytes > ds->ds_quota) {
2782 		dsl_dataset_rele(ds, FTAG);
2783 		return (SET_ERROR(EDQUOT));
2784 	}
2785 
2786 	/*
2787 	 * When we do the clone swap, we will temporarily use more space
2788 	 * due to the refreservation (the head will no longer have any
2789 	 * unique space, so the entire amount of the refreservation will need
2790 	 * to be free).  We will immediately destroy the clone, freeing
2791 	 * this space, but the freeing happens over many txg's.
2792 	 */
2793 	unused_refres_delta = (int64_t)MIN(ds->ds_reserved,
2794 	    dsl_dataset_phys(ds)->ds_unique_bytes);
2795 
2796 	if (unused_refres_delta > 0 &&
2797 	    unused_refres_delta >
2798 	    dsl_dir_space_available(ds->ds_dir, NULL, 0, TRUE)) {
2799 		dsl_dataset_rele(ds, FTAG);
2800 		return (SET_ERROR(ENOSPC));
2801 	}
2802 
2803 	dsl_dataset_rele(ds, FTAG);
2804 	return (0);
2805 }
2806 
2807 void
dsl_dataset_rollback_sync(void * arg,dmu_tx_t * tx)2808 dsl_dataset_rollback_sync(void *arg, dmu_tx_t *tx)
2809 {
2810 	dsl_dataset_rollback_arg_t *ddra = arg;
2811 	dsl_pool_t *dp = dmu_tx_pool(tx);
2812 	dsl_dataset_t *ds, *clone;
2813 	uint64_t cloneobj;
2814 	char namebuf[ZFS_MAX_DATASET_NAME_LEN];
2815 
2816 	VERIFY0(dsl_dataset_hold(dp, ddra->ddra_fsname, FTAG, &ds));
2817 
2818 	dsl_dataset_name(ds->ds_prev, namebuf);
2819 	fnvlist_add_string(ddra->ddra_result, "target", namebuf);
2820 
2821 	cloneobj = dsl_dataset_create_sync(ds->ds_dir, "%rollback",
2822 	    ds->ds_prev, DS_CREATE_FLAG_NODIRTY, kcred, NULL, tx);
2823 
2824 	VERIFY0(dsl_dataset_hold_obj(dp, cloneobj, FTAG, &clone));
2825 
2826 	dsl_dataset_clone_swap_sync_impl(clone, ds, tx);
2827 	dsl_dataset_zero_zil(ds, tx);
2828 
2829 	dsl_destroy_head_sync_impl(clone, tx);
2830 
2831 	dsl_dataset_rele(clone, FTAG);
2832 	dsl_dataset_rele(ds, FTAG);
2833 }
2834 
2835 /*
2836  * Rolls back the given filesystem or volume to the most recent snapshot.
2837  * The name of the most recent snapshot will be returned under key "target"
2838  * in the result nvlist.
2839  *
2840  * If owner != NULL:
2841  * - The existing dataset MUST be owned by the specified owner at entry
2842  * - Upon return, dataset will still be held by the same owner, whether we
2843  *   succeed or not.
2844  *
2845  * This mode is required any time the existing filesystem is mounted.  See
2846  * notes above zfs_suspend_fs() for further details.
2847  */
2848 int
dsl_dataset_rollback(const char * fsname,const char * tosnap,void * owner,nvlist_t * result)2849 dsl_dataset_rollback(const char *fsname, const char *tosnap, void *owner,
2850     nvlist_t *result)
2851 {
2852 	dsl_dataset_rollback_arg_t ddra;
2853 
2854 	ddra.ddra_fsname = fsname;
2855 	ddra.ddra_tosnap = tosnap;
2856 	ddra.ddra_owner = owner;
2857 	ddra.ddra_result = result;
2858 
2859 	return (dsl_sync_task(fsname, dsl_dataset_rollback_check,
2860 	    dsl_dataset_rollback_sync, &ddra,
2861 	    1, ZFS_SPACE_CHECK_RESERVED));
2862 }
2863 
2864 struct promotenode {
2865 	list_node_t link;
2866 	dsl_dataset_t *ds;
2867 };
2868 
2869 static int snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep);
2870 static int promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp,
2871     void *tag);
2872 static void promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag);
2873 
2874 int
dsl_dataset_promote_check(void * arg,dmu_tx_t * tx)2875 dsl_dataset_promote_check(void *arg, dmu_tx_t *tx)
2876 {
2877 	dsl_dataset_promote_arg_t *ddpa = arg;
2878 	dsl_pool_t *dp = dmu_tx_pool(tx);
2879 	dsl_dataset_t *hds;
2880 	struct promotenode *snap;
2881 	dsl_dataset_t *origin_ds;
2882 	int err;
2883 	uint64_t unused;
2884 	uint64_t ss_mv_cnt;
2885 	size_t max_snap_len;
2886 	boolean_t conflicting_snaps;
2887 
2888 	err = promote_hold(ddpa, dp, FTAG);
2889 	if (err != 0)
2890 		return (err);
2891 
2892 	hds = ddpa->ddpa_clone;
2893 	snap = list_head(&ddpa->shared_snaps);
2894 	origin_ds = snap->ds;
2895 	max_snap_len = MAXNAMELEN - strlen(ddpa->ddpa_clonename) - 1;
2896 
2897 	snap = list_head(&ddpa->origin_snaps);
2898 
2899 	if (dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE) {
2900 		promote_rele(ddpa, FTAG);
2901 		return (SET_ERROR(EXDEV));
2902 	}
2903 
2904 	snap = list_head(&ddpa->shared_snaps);
2905 	if (snap == NULL) {
2906 		err = SET_ERROR(ENOENT);
2907 		goto out;
2908 	}
2909 	origin_ds = snap->ds;
2910 
2911 	/*
2912 	 * Encrypted clones share a DSL Crypto Key with their origin's dsl dir.
2913 	 * When doing a promote we must make sure the encryption root for
2914 	 * both the target and the target's origin does not change to avoid
2915 	 * needing to rewrap encryption keys
2916 	 */
2917 	err = dsl_dataset_promote_crypt_check(hds->ds_dir, origin_ds->ds_dir);
2918 	if (err != 0)
2919 		goto out;
2920 
2921 	/*
2922 	 * Compute and check the amount of space to transfer.  Since this is
2923 	 * so expensive, don't do the preliminary check.
2924 	 */
2925 	if (!dmu_tx_is_syncing(tx)) {
2926 		promote_rele(ddpa, FTAG);
2927 		return (0);
2928 	}
2929 
2930 	/* compute origin's new unique space */
2931 	snap = list_tail(&ddpa->clone_snaps);
2932 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
2933 	    origin_ds->ds_object);
2934 	dsl_deadlist_space_range(&snap->ds->ds_deadlist,
2935 	    dsl_dataset_phys(origin_ds)->ds_prev_snap_txg, UINT64_MAX,
2936 	    &ddpa->unique, &unused, &unused);
2937 
2938 	/*
2939 	 * Walk the snapshots that we are moving
2940 	 *
2941 	 * Compute space to transfer.  Consider the incremental changes
2942 	 * to used by each snapshot:
2943 	 * (my used) = (prev's used) + (blocks born) - (blocks killed)
2944 	 * So each snapshot gave birth to:
2945 	 * (blocks born) = (my used) - (prev's used) + (blocks killed)
2946 	 * So a sequence would look like:
2947 	 * (uN - u(N-1) + kN) + ... + (u1 - u0 + k1) + (u0 - 0 + k0)
2948 	 * Which simplifies to:
2949 	 * uN + kN + kN-1 + ... + k1 + k0
2950 	 * Note however, if we stop before we reach the ORIGIN we get:
2951 	 * uN + kN + kN-1 + ... + kM - uM-1
2952 	 */
2953 	conflicting_snaps = B_FALSE;
2954 	ss_mv_cnt = 0;
2955 	ddpa->used = dsl_dataset_phys(origin_ds)->ds_referenced_bytes;
2956 	ddpa->comp = dsl_dataset_phys(origin_ds)->ds_compressed_bytes;
2957 	ddpa->uncomp = dsl_dataset_phys(origin_ds)->ds_uncompressed_bytes;
2958 	for (snap = list_head(&ddpa->shared_snaps); snap;
2959 	    snap = list_next(&ddpa->shared_snaps, snap)) {
2960 		uint64_t val, dlused, dlcomp, dluncomp;
2961 		dsl_dataset_t *ds = snap->ds;
2962 
2963 		ss_mv_cnt++;
2964 
2965 		/*
2966 		 * If there are long holds, we won't be able to evict
2967 		 * the objset.
2968 		 */
2969 		if (dsl_dataset_long_held(ds)) {
2970 			err = SET_ERROR(EBUSY);
2971 			goto out;
2972 		}
2973 
2974 		/* Check that the snapshot name does not conflict */
2975 		VERIFY0(dsl_dataset_get_snapname(ds));
2976 		if (strlen(ds->ds_snapname) >= max_snap_len) {
2977 			err = SET_ERROR(ENAMETOOLONG);
2978 			goto out;
2979 		}
2980 		err = dsl_dataset_snap_lookup(hds, ds->ds_snapname, &val);
2981 		if (err == 0) {
2982 			fnvlist_add_boolean(ddpa->err_ds,
2983 			    snap->ds->ds_snapname);
2984 			conflicting_snaps = B_TRUE;
2985 		} else if (err != ENOENT) {
2986 			goto out;
2987 		}
2988 
2989 		/* The very first snapshot does not have a deadlist */
2990 		if (dsl_dataset_phys(ds)->ds_prev_snap_obj == 0)
2991 			continue;
2992 
2993 		dsl_deadlist_space(&ds->ds_deadlist,
2994 		    &dlused, &dlcomp, &dluncomp);
2995 		ddpa->used += dlused;
2996 		ddpa->comp += dlcomp;
2997 		ddpa->uncomp += dluncomp;
2998 	}
2999 
3000 	/*
3001 	 * In order to return the full list of conflicting snapshots, we check
3002 	 * whether there was a conflict after traversing all of them.
3003 	 */
3004 	if (conflicting_snaps) {
3005 		err = SET_ERROR(EEXIST);
3006 		goto out;
3007 	}
3008 
3009 	/*
3010 	 * If we are a clone of a clone then we never reached ORIGIN,
3011 	 * so we need to subtract out the clone origin's used space.
3012 	 */
3013 	if (ddpa->origin_origin) {
3014 		ddpa->used -=
3015 		    dsl_dataset_phys(ddpa->origin_origin)->ds_referenced_bytes;
3016 		ddpa->comp -=
3017 		    dsl_dataset_phys(ddpa->origin_origin)->ds_compressed_bytes;
3018 		ddpa->uncomp -=
3019 		    dsl_dataset_phys(ddpa->origin_origin)->
3020 		    ds_uncompressed_bytes;
3021 	}
3022 
3023 	/* Check that there is enough space and limit headroom here */
3024 	err = dsl_dir_transfer_possible(origin_ds->ds_dir, hds->ds_dir,
3025 	    0, ss_mv_cnt, ddpa->used, ddpa->cr);
3026 	if (err != 0)
3027 		goto out;
3028 
3029 	/*
3030 	 * Compute the amounts of space that will be used by snapshots
3031 	 * after the promotion (for both origin and clone).  For each,
3032 	 * it is the amount of space that will be on all of their
3033 	 * deadlists (that was not born before their new origin).
3034 	 */
3035 	if (dsl_dir_phys(hds->ds_dir)->dd_flags & DD_FLAG_USED_BREAKDOWN) {
3036 		uint64_t space;
3037 
3038 		/*
3039 		 * Note, typically this will not be a clone of a clone,
3040 		 * so dd_origin_txg will be < TXG_INITIAL, so
3041 		 * these snaplist_space() -> dsl_deadlist_space_range()
3042 		 * calls will be fast because they do not have to
3043 		 * iterate over all bps.
3044 		 */
3045 		snap = list_head(&ddpa->origin_snaps);
3046 		err = snaplist_space(&ddpa->shared_snaps,
3047 		    snap->ds->ds_dir->dd_origin_txg, &ddpa->cloneusedsnap);
3048 		if (err != 0)
3049 			goto out;
3050 
3051 		err = snaplist_space(&ddpa->clone_snaps,
3052 		    snap->ds->ds_dir->dd_origin_txg, &space);
3053 		if (err != 0)
3054 			goto out;
3055 		ddpa->cloneusedsnap += space;
3056 	}
3057 	if (dsl_dir_phys(origin_ds->ds_dir)->dd_flags &
3058 	    DD_FLAG_USED_BREAKDOWN) {
3059 		err = snaplist_space(&ddpa->origin_snaps,
3060 		    dsl_dataset_phys(origin_ds)->ds_creation_txg,
3061 		    &ddpa->originusedsnap);
3062 		if (err != 0)
3063 			goto out;
3064 	}
3065 
3066 out:
3067 	promote_rele(ddpa, FTAG);
3068 	return (err);
3069 }
3070 
3071 void
dsl_dataset_promote_sync(void * arg,dmu_tx_t * tx)3072 dsl_dataset_promote_sync(void *arg, dmu_tx_t *tx)
3073 {
3074 	dsl_dataset_promote_arg_t *ddpa = arg;
3075 	dsl_pool_t *dp = dmu_tx_pool(tx);
3076 	dsl_dataset_t *hds;
3077 	struct promotenode *snap;
3078 	dsl_dataset_t *origin_ds;
3079 	dsl_dataset_t *origin_head;
3080 	dsl_dir_t *dd;
3081 	dsl_dir_t *odd = NULL;
3082 	uint64_t oldnext_obj;
3083 	int64_t delta;
3084 
3085 	VERIFY0(promote_hold(ddpa, dp, FTAG));
3086 	hds = ddpa->ddpa_clone;
3087 
3088 	ASSERT0(dsl_dataset_phys(hds)->ds_flags & DS_FLAG_NOPROMOTE);
3089 
3090 	snap = list_head(&ddpa->shared_snaps);
3091 	origin_ds = snap->ds;
3092 	dd = hds->ds_dir;
3093 
3094 	snap = list_head(&ddpa->origin_snaps);
3095 	origin_head = snap->ds;
3096 
3097 	/*
3098 	 * We need to explicitly open odd, since origin_ds's dd will be
3099 	 * changing.
3100 	 */
3101 	VERIFY0(dsl_dir_hold_obj(dp, origin_ds->ds_dir->dd_object,
3102 	    NULL, FTAG, &odd));
3103 
3104 	dsl_dataset_promote_crypt_sync(hds->ds_dir, odd, tx);
3105 
3106 	/* change origin's next snap */
3107 	dmu_buf_will_dirty(origin_ds->ds_dbuf, tx);
3108 	oldnext_obj = dsl_dataset_phys(origin_ds)->ds_next_snap_obj;
3109 	snap = list_tail(&ddpa->clone_snaps);
3110 	ASSERT3U(dsl_dataset_phys(snap->ds)->ds_prev_snap_obj, ==,
3111 	    origin_ds->ds_object);
3112 	dsl_dataset_phys(origin_ds)->ds_next_snap_obj = snap->ds->ds_object;
3113 
3114 	/* change the origin's next clone */
3115 	if (dsl_dataset_phys(origin_ds)->ds_next_clones_obj) {
3116 		dsl_dataset_remove_from_next_clones(origin_ds,
3117 		    snap->ds->ds_object, tx);
3118 		VERIFY0(zap_add_int(dp->dp_meta_objset,
3119 		    dsl_dataset_phys(origin_ds)->ds_next_clones_obj,
3120 		    oldnext_obj, tx));
3121 	}
3122 
3123 	/* change origin */
3124 	dmu_buf_will_dirty(dd->dd_dbuf, tx);
3125 	ASSERT3U(dsl_dir_phys(dd)->dd_origin_obj, ==, origin_ds->ds_object);
3126 	dsl_dir_phys(dd)->dd_origin_obj = dsl_dir_phys(odd)->dd_origin_obj;
3127 	dd->dd_origin_txg = origin_head->ds_dir->dd_origin_txg;
3128 	dmu_buf_will_dirty(odd->dd_dbuf, tx);
3129 	dsl_dir_phys(odd)->dd_origin_obj = origin_ds->ds_object;
3130 	origin_head->ds_dir->dd_origin_txg =
3131 	    dsl_dataset_phys(origin_ds)->ds_creation_txg;
3132 
3133 	/* change dd_clone entries */
3134 	if (spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
3135 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
3136 		    dsl_dir_phys(odd)->dd_clones, hds->ds_object, tx));
3137 		VERIFY0(zap_add_int(dp->dp_meta_objset,
3138 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
3139 		    hds->ds_object, tx));
3140 
3141 		VERIFY0(zap_remove_int(dp->dp_meta_objset,
3142 		    dsl_dir_phys(ddpa->origin_origin->ds_dir)->dd_clones,
3143 		    origin_head->ds_object, tx));
3144 		if (dsl_dir_phys(dd)->dd_clones == 0) {
3145 			dsl_dir_phys(dd)->dd_clones =
3146 			    zap_create(dp->dp_meta_objset, DMU_OT_DSL_CLONES,
3147 			    DMU_OT_NONE, 0, tx);
3148 		}
3149 		VERIFY0(zap_add_int(dp->dp_meta_objset,
3150 		    dsl_dir_phys(dd)->dd_clones, origin_head->ds_object, tx));
3151 	}
3152 
3153 	/* move snapshots to this dir */
3154 	for (snap = list_head(&ddpa->shared_snaps); snap;
3155 	    snap = list_next(&ddpa->shared_snaps, snap)) {
3156 		dsl_dataset_t *ds = snap->ds;
3157 
3158 		/*
3159 		 * Property callbacks are registered to a particular
3160 		 * dsl_dir.  Since ours is changing, evict the objset
3161 		 * so that they will be unregistered from the old dsl_dir.
3162 		 */
3163 		if (ds->ds_objset) {
3164 			dmu_objset_evict(ds->ds_objset);
3165 			ds->ds_objset = NULL;
3166 		}
3167 
3168 		/* move snap name entry */
3169 		VERIFY0(dsl_dataset_get_snapname(ds));
3170 		VERIFY0(dsl_dataset_snap_remove(origin_head,
3171 		    ds->ds_snapname, tx, B_TRUE));
3172 		VERIFY0(zap_add(dp->dp_meta_objset,
3173 		    dsl_dataset_phys(hds)->ds_snapnames_zapobj, ds->ds_snapname,
3174 		    8, 1, &ds->ds_object, tx));
3175 		dsl_fs_ss_count_adjust(hds->ds_dir, 1,
3176 		    DD_FIELD_SNAPSHOT_COUNT, tx);
3177 
3178 		/* change containing dsl_dir */
3179 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3180 		ASSERT3U(dsl_dataset_phys(ds)->ds_dir_obj, ==, odd->dd_object);
3181 		dsl_dataset_phys(ds)->ds_dir_obj = dd->dd_object;
3182 		ASSERT3P(ds->ds_dir, ==, odd);
3183 		dsl_dir_rele(ds->ds_dir, ds);
3184 		VERIFY0(dsl_dir_hold_obj(dp, dd->dd_object,
3185 		    NULL, ds, &ds->ds_dir));
3186 
3187 		/* move any clone references */
3188 		if (dsl_dataset_phys(ds)->ds_next_clones_obj &&
3189 		    spa_version(dp->dp_spa) >= SPA_VERSION_DIR_CLONES) {
3190 			zap_cursor_t zc;
3191 			zap_attribute_t za;
3192 
3193 			for (zap_cursor_init(&zc, dp->dp_meta_objset,
3194 			    dsl_dataset_phys(ds)->ds_next_clones_obj);
3195 			    zap_cursor_retrieve(&zc, &za) == 0;
3196 			    zap_cursor_advance(&zc)) {
3197 				dsl_dataset_t *cnds;
3198 				uint64_t o;
3199 
3200 				if (za.za_first_integer == oldnext_obj) {
3201 					/*
3202 					 * We've already moved the
3203 					 * origin's reference.
3204 					 */
3205 					continue;
3206 				}
3207 
3208 				VERIFY0(dsl_dataset_hold_obj(dp,
3209 				    za.za_first_integer, FTAG, &cnds));
3210 				o = dsl_dir_phys(cnds->ds_dir)->
3211 				    dd_head_dataset_obj;
3212 
3213 				VERIFY0(zap_remove_int(dp->dp_meta_objset,
3214 				    dsl_dir_phys(odd)->dd_clones, o, tx));
3215 				VERIFY0(zap_add_int(dp->dp_meta_objset,
3216 				    dsl_dir_phys(dd)->dd_clones, o, tx));
3217 				dsl_dataset_rele(cnds, FTAG);
3218 			}
3219 			zap_cursor_fini(&zc);
3220 		}
3221 
3222 		ASSERT(!dsl_prop_hascb(ds));
3223 	}
3224 
3225 	/*
3226 	 * Change space accounting.
3227 	 * Note, pa->*usedsnap and dd_used_breakdown[SNAP] will either
3228 	 * both be valid, or both be 0 (resulting in delta == 0).  This
3229 	 * is true for each of {clone,origin} independently.
3230 	 */
3231 
3232 	delta = ddpa->cloneusedsnap -
3233 	    dsl_dir_phys(dd)->dd_used_breakdown[DD_USED_SNAP];
3234 	ASSERT3S(delta, >=, 0);
3235 	ASSERT3U(ddpa->used, >=, delta);
3236 	dsl_dir_diduse_space(dd, DD_USED_SNAP, delta, 0, 0, tx);
3237 	dsl_dir_diduse_space(dd, DD_USED_HEAD,
3238 	    ddpa->used - delta, ddpa->comp, ddpa->uncomp, tx);
3239 
3240 	delta = ddpa->originusedsnap -
3241 	    dsl_dir_phys(odd)->dd_used_breakdown[DD_USED_SNAP];
3242 	ASSERT3S(delta, <=, 0);
3243 	ASSERT3U(ddpa->used, >=, -delta);
3244 	dsl_dir_diduse_space(odd, DD_USED_SNAP, delta, 0, 0, tx);
3245 	dsl_dir_diduse_space(odd, DD_USED_HEAD,
3246 	    -ddpa->used - delta, -ddpa->comp, -ddpa->uncomp, tx);
3247 
3248 	dsl_dataset_phys(origin_ds)->ds_unique_bytes = ddpa->unique;
3249 
3250 	/* log history record */
3251 	spa_history_log_internal_ds(hds, "promote", tx, "");
3252 
3253 	dsl_dir_rele(odd, FTAG);
3254 	promote_rele(ddpa, FTAG);
3255 }
3256 
3257 /*
3258  * Make a list of dsl_dataset_t's for the snapshots between first_obj
3259  * (exclusive) and last_obj (inclusive).  The list will be in reverse
3260  * order (last_obj will be the list_head()).  If first_obj == 0, do all
3261  * snapshots back to this dataset's origin.
3262  */
3263 static int
snaplist_make(dsl_pool_t * dp,uint64_t first_obj,uint64_t last_obj,list_t * l,void * tag)3264 snaplist_make(dsl_pool_t *dp,
3265     uint64_t first_obj, uint64_t last_obj, list_t *l, void *tag)
3266 {
3267 	uint64_t obj = last_obj;
3268 
3269 	list_create(l, sizeof (struct promotenode),
3270 	    offsetof(struct promotenode, link));
3271 
3272 	while (obj != first_obj) {
3273 		dsl_dataset_t *ds;
3274 		struct promotenode *snap;
3275 		int err;
3276 
3277 		err = dsl_dataset_hold_obj(dp, obj, tag, &ds);
3278 		ASSERT(err != ENOENT);
3279 		if (err != 0)
3280 			return (err);
3281 
3282 		if (first_obj == 0)
3283 			first_obj = dsl_dir_phys(ds->ds_dir)->dd_origin_obj;
3284 
3285 		snap = kmem_alloc(sizeof (*snap), KM_SLEEP);
3286 		snap->ds = ds;
3287 		list_insert_tail(l, snap);
3288 		obj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
3289 	}
3290 
3291 	return (0);
3292 }
3293 
3294 static int
snaplist_space(list_t * l,uint64_t mintxg,uint64_t * spacep)3295 snaplist_space(list_t *l, uint64_t mintxg, uint64_t *spacep)
3296 {
3297 	struct promotenode *snap;
3298 
3299 	*spacep = 0;
3300 	for (snap = list_head(l); snap; snap = list_next(l, snap)) {
3301 		uint64_t used, comp, uncomp;
3302 		dsl_deadlist_space_range(&snap->ds->ds_deadlist,
3303 		    mintxg, UINT64_MAX, &used, &comp, &uncomp);
3304 		*spacep += used;
3305 	}
3306 	return (0);
3307 }
3308 
3309 static void
snaplist_destroy(list_t * l,void * tag)3310 snaplist_destroy(list_t *l, void *tag)
3311 {
3312 	struct promotenode *snap;
3313 
3314 	if (l == NULL || !list_link_active(&l->list_head))
3315 		return;
3316 
3317 	while ((snap = list_tail(l)) != NULL) {
3318 		list_remove(l, snap);
3319 		dsl_dataset_rele(snap->ds, tag);
3320 		kmem_free(snap, sizeof (*snap));
3321 	}
3322 	list_destroy(l);
3323 }
3324 
3325 static int
promote_hold(dsl_dataset_promote_arg_t * ddpa,dsl_pool_t * dp,void * tag)3326 promote_hold(dsl_dataset_promote_arg_t *ddpa, dsl_pool_t *dp, void *tag)
3327 {
3328 	int error;
3329 	dsl_dir_t *dd;
3330 	struct promotenode *snap;
3331 
3332 	error = dsl_dataset_hold(dp, ddpa->ddpa_clonename, tag,
3333 	    &ddpa->ddpa_clone);
3334 	if (error != 0)
3335 		return (error);
3336 	dd = ddpa->ddpa_clone->ds_dir;
3337 
3338 	if (ddpa->ddpa_clone->ds_is_snapshot ||
3339 	    !dsl_dir_is_clone(dd)) {
3340 		dsl_dataset_rele(ddpa->ddpa_clone, tag);
3341 		return (SET_ERROR(EINVAL));
3342 	}
3343 
3344 	error = snaplist_make(dp, 0, dsl_dir_phys(dd)->dd_origin_obj,
3345 	    &ddpa->shared_snaps, tag);
3346 	if (error != 0)
3347 		goto out;
3348 
3349 	error = snaplist_make(dp, 0, ddpa->ddpa_clone->ds_object,
3350 	    &ddpa->clone_snaps, tag);
3351 	if (error != 0)
3352 		goto out;
3353 
3354 	snap = list_head(&ddpa->shared_snaps);
3355 	ASSERT3U(snap->ds->ds_object, ==, dsl_dir_phys(dd)->dd_origin_obj);
3356 	error = snaplist_make(dp, dsl_dir_phys(dd)->dd_origin_obj,
3357 	    dsl_dir_phys(snap->ds->ds_dir)->dd_head_dataset_obj,
3358 	    &ddpa->origin_snaps, tag);
3359 	if (error != 0)
3360 		goto out;
3361 
3362 	if (dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj != 0) {
3363 		error = dsl_dataset_hold_obj(dp,
3364 		    dsl_dir_phys(snap->ds->ds_dir)->dd_origin_obj,
3365 		    tag, &ddpa->origin_origin);
3366 		if (error != 0)
3367 			goto out;
3368 	}
3369 out:
3370 	if (error != 0)
3371 		promote_rele(ddpa, tag);
3372 	return (error);
3373 }
3374 
3375 static void
promote_rele(dsl_dataset_promote_arg_t * ddpa,void * tag)3376 promote_rele(dsl_dataset_promote_arg_t *ddpa, void *tag)
3377 {
3378 	snaplist_destroy(&ddpa->shared_snaps, tag);
3379 	snaplist_destroy(&ddpa->clone_snaps, tag);
3380 	snaplist_destroy(&ddpa->origin_snaps, tag);
3381 	if (ddpa->origin_origin != NULL)
3382 		dsl_dataset_rele(ddpa->origin_origin, tag);
3383 	dsl_dataset_rele(ddpa->ddpa_clone, tag);
3384 }
3385 
3386 /*
3387  * Promote a clone.
3388  *
3389  * If it fails due to a conflicting snapshot name, "conflsnap" will be filled
3390  * in with the name.  (It must be at least ZFS_MAX_DATASET_NAME_LEN bytes long.)
3391  */
3392 int
dsl_dataset_promote(const char * name,char * conflsnap)3393 dsl_dataset_promote(const char *name, char *conflsnap)
3394 {
3395 	dsl_dataset_promote_arg_t ddpa = { 0 };
3396 	uint64_t numsnaps;
3397 	int error;
3398 	nvpair_t *snap_pair;
3399 	objset_t *os;
3400 
3401 	/*
3402 	 * We will modify space proportional to the number of
3403 	 * snapshots.  Compute numsnaps.
3404 	 */
3405 	error = dmu_objset_hold(name, FTAG, &os);
3406 	if (error != 0)
3407 		return (error);
3408 	error = zap_count(dmu_objset_pool(os)->dp_meta_objset,
3409 	    dsl_dataset_phys(dmu_objset_ds(os))->ds_snapnames_zapobj,
3410 	    &numsnaps);
3411 	dmu_objset_rele(os, FTAG);
3412 	if (error != 0)
3413 		return (error);
3414 
3415 	ddpa.ddpa_clonename = name;
3416 	ddpa.err_ds = fnvlist_alloc();
3417 	ddpa.cr = CRED();
3418 
3419 	error = dsl_sync_task(name, dsl_dataset_promote_check,
3420 	    dsl_dataset_promote_sync, &ddpa,
3421 	    2 + numsnaps, ZFS_SPACE_CHECK_RESERVED);
3422 
3423 	/*
3424 	 * Return the first conflicting snapshot found.
3425 	 */
3426 	snap_pair = nvlist_next_nvpair(ddpa.err_ds, NULL);
3427 	if (snap_pair != NULL && conflsnap != NULL)
3428 		(void) strcpy(conflsnap, nvpair_name(snap_pair));
3429 
3430 	fnvlist_free(ddpa.err_ds);
3431 	return (error);
3432 }
3433 
3434 int
dsl_dataset_clone_swap_check_impl(dsl_dataset_t * clone,dsl_dataset_t * origin_head,boolean_t force,void * owner,dmu_tx_t * tx)3435 dsl_dataset_clone_swap_check_impl(dsl_dataset_t *clone,
3436     dsl_dataset_t *origin_head, boolean_t force, void *owner, dmu_tx_t *tx)
3437 {
3438 	/*
3439 	 * "slack" factor for received datasets with refquota set on them.
3440 	 * See the bottom of this function for details on its use.
3441 	 */
3442 	uint64_t refquota_slack = DMU_MAX_ACCESS * spa_asize_inflation;
3443 	int64_t unused_refres_delta;
3444 
3445 	/* they should both be heads */
3446 	if (clone->ds_is_snapshot ||
3447 	    origin_head->ds_is_snapshot)
3448 		return (SET_ERROR(EINVAL));
3449 
3450 	/* if we are not forcing, the branch point should be just before them */
3451 	if (!force && clone->ds_prev != origin_head->ds_prev)
3452 		return (SET_ERROR(EINVAL));
3453 
3454 	/* clone should be the clone (unless they are unrelated) */
3455 	if (clone->ds_prev != NULL &&
3456 	    clone->ds_prev != clone->ds_dir->dd_pool->dp_origin_snap &&
3457 	    origin_head->ds_dir != clone->ds_prev->ds_dir)
3458 		return (SET_ERROR(EINVAL));
3459 
3460 	/* the clone should be a child of the origin */
3461 	if (clone->ds_dir->dd_parent != origin_head->ds_dir)
3462 		return (SET_ERROR(EINVAL));
3463 
3464 	/* origin_head shouldn't be modified unless 'force' */
3465 	if (!force &&
3466 	    dsl_dataset_modified_since_snap(origin_head, origin_head->ds_prev))
3467 		return (SET_ERROR(ETXTBSY));
3468 
3469 	/* origin_head should have no long holds (e.g. is not mounted) */
3470 	if (dsl_dataset_handoff_check(origin_head, owner, tx))
3471 		return (SET_ERROR(EBUSY));
3472 
3473 	/* check amount of any unconsumed refreservation */
3474 	unused_refres_delta =
3475 	    (int64_t)MIN(origin_head->ds_reserved,
3476 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3477 	    (int64_t)MIN(origin_head->ds_reserved,
3478 	    dsl_dataset_phys(clone)->ds_unique_bytes);
3479 
3480 	if (unused_refres_delta > 0 &&
3481 	    unused_refres_delta >
3482 	    dsl_dir_space_available(origin_head->ds_dir, NULL, 0, TRUE))
3483 		return (SET_ERROR(ENOSPC));
3484 
3485 	/*
3486 	 * The clone can't be too much over the head's refquota.
3487 	 *
3488 	 * To ensure that the entire refquota can be used, we allow one
3489 	 * transaction to exceed the the refquota.  Therefore, this check
3490 	 * needs to also allow for the space referenced to be more than the
3491 	 * refquota.  The maximum amount of space that one transaction can use
3492 	 * on disk is DMU_MAX_ACCESS * spa_asize_inflation.  Allowing this
3493 	 * overage ensures that we are able to receive a filesystem that
3494 	 * exceeds the refquota on the source system.
3495 	 *
3496 	 * So that overage is the refquota_slack we use below.
3497 	 */
3498 	if (origin_head->ds_quota != 0 &&
3499 	    dsl_dataset_phys(clone)->ds_referenced_bytes >
3500 	    origin_head->ds_quota + refquota_slack)
3501 		return (SET_ERROR(EDQUOT));
3502 
3503 	return (0);
3504 }
3505 
3506 static void
dsl_dataset_swap_remap_deadlists(dsl_dataset_t * clone,dsl_dataset_t * origin,dmu_tx_t * tx)3507 dsl_dataset_swap_remap_deadlists(dsl_dataset_t *clone,
3508     dsl_dataset_t *origin, dmu_tx_t *tx)
3509 {
3510 	uint64_t clone_remap_dl_obj, origin_remap_dl_obj;
3511 	dsl_pool_t *dp = dmu_tx_pool(tx);
3512 
3513 	ASSERT(dsl_pool_sync_context(dp));
3514 
3515 	clone_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(clone);
3516 	origin_remap_dl_obj = dsl_dataset_get_remap_deadlist_object(origin);
3517 
3518 	if (clone_remap_dl_obj != 0) {
3519 		dsl_deadlist_close(&clone->ds_remap_deadlist);
3520 		dsl_dataset_unset_remap_deadlist_object(clone, tx);
3521 	}
3522 	if (origin_remap_dl_obj != 0) {
3523 		dsl_deadlist_close(&origin->ds_remap_deadlist);
3524 		dsl_dataset_unset_remap_deadlist_object(origin, tx);
3525 	}
3526 
3527 	if (clone_remap_dl_obj != 0) {
3528 		dsl_dataset_set_remap_deadlist_object(origin,
3529 		    clone_remap_dl_obj, tx);
3530 		dsl_deadlist_open(&origin->ds_remap_deadlist,
3531 		    dp->dp_meta_objset, clone_remap_dl_obj);
3532 	}
3533 	if (origin_remap_dl_obj != 0) {
3534 		dsl_dataset_set_remap_deadlist_object(clone,
3535 		    origin_remap_dl_obj, tx);
3536 		dsl_deadlist_open(&clone->ds_remap_deadlist,
3537 		    dp->dp_meta_objset, origin_remap_dl_obj);
3538 	}
3539 }
3540 
3541 void
dsl_dataset_clone_swap_sync_impl(dsl_dataset_t * clone,dsl_dataset_t * origin_head,dmu_tx_t * tx)3542 dsl_dataset_clone_swap_sync_impl(dsl_dataset_t *clone,
3543     dsl_dataset_t *origin_head, dmu_tx_t *tx)
3544 {
3545 	dsl_pool_t *dp = dmu_tx_pool(tx);
3546 	int64_t unused_refres_delta;
3547 
3548 	ASSERT(clone->ds_reserved == 0);
3549 	/*
3550 	 * NOTE: On DEBUG kernels there could be a race between this and
3551 	 * the check function if spa_asize_inflation is adjusted...
3552 	 */
3553 	ASSERT(origin_head->ds_quota == 0 ||
3554 	    dsl_dataset_phys(clone)->ds_unique_bytes <= origin_head->ds_quota +
3555 	    DMU_MAX_ACCESS * spa_asize_inflation);
3556 	ASSERT3P(clone->ds_prev, ==, origin_head->ds_prev);
3557 
3558 	/*
3559 	 * Swap per-dataset feature flags.
3560 	 */
3561 	for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
3562 		if (!(spa_feature_table[f].fi_flags &
3563 		    ZFEATURE_FLAG_PER_DATASET)) {
3564 			ASSERT(!clone->ds_feature_inuse[f]);
3565 			ASSERT(!origin_head->ds_feature_inuse[f]);
3566 			continue;
3567 		}
3568 
3569 		boolean_t clone_inuse = clone->ds_feature_inuse[f];
3570 		boolean_t origin_head_inuse = origin_head->ds_feature_inuse[f];
3571 
3572 		if (clone_inuse) {
3573 			dsl_dataset_deactivate_feature(clone->ds_object, f, tx);
3574 			clone->ds_feature_inuse[f] = B_FALSE;
3575 		}
3576 		if (origin_head_inuse) {
3577 			dsl_dataset_deactivate_feature(origin_head->ds_object,
3578 			    f, tx);
3579 			origin_head->ds_feature_inuse[f] = B_FALSE;
3580 		}
3581 		if (clone_inuse) {
3582 			dsl_dataset_activate_feature(origin_head->ds_object,
3583 			    f, tx);
3584 			origin_head->ds_feature_inuse[f] = B_TRUE;
3585 		}
3586 		if (origin_head_inuse) {
3587 			dsl_dataset_activate_feature(clone->ds_object, f, tx);
3588 			clone->ds_feature_inuse[f] = B_TRUE;
3589 		}
3590 	}
3591 
3592 	dmu_buf_will_dirty(clone->ds_dbuf, tx);
3593 	dmu_buf_will_dirty(origin_head->ds_dbuf, tx);
3594 
3595 	if (clone->ds_objset != NULL) {
3596 		dmu_objset_evict(clone->ds_objset);
3597 		clone->ds_objset = NULL;
3598 	}
3599 
3600 	if (origin_head->ds_objset != NULL) {
3601 		dmu_objset_evict(origin_head->ds_objset);
3602 		origin_head->ds_objset = NULL;
3603 	}
3604 
3605 	unused_refres_delta =
3606 	    (int64_t)MIN(origin_head->ds_reserved,
3607 	    dsl_dataset_phys(origin_head)->ds_unique_bytes) -
3608 	    (int64_t)MIN(origin_head->ds_reserved,
3609 	    dsl_dataset_phys(clone)->ds_unique_bytes);
3610 
3611 	/*
3612 	 * Reset origin's unique bytes, if it exists.
3613 	 */
3614 	if (clone->ds_prev) {
3615 		dsl_dataset_t *origin = clone->ds_prev;
3616 		uint64_t comp, uncomp;
3617 
3618 		dmu_buf_will_dirty(origin->ds_dbuf, tx);
3619 		dsl_deadlist_space_range(&clone->ds_deadlist,
3620 		    dsl_dataset_phys(origin)->ds_prev_snap_txg, UINT64_MAX,
3621 		    &dsl_dataset_phys(origin)->ds_unique_bytes, &comp, &uncomp);
3622 	}
3623 
3624 	/* swap blkptrs */
3625 	{
3626 		rrw_enter(&clone->ds_bp_rwlock, RW_WRITER, FTAG);
3627 		rrw_enter(&origin_head->ds_bp_rwlock, RW_WRITER, FTAG);
3628 		blkptr_t tmp;
3629 		tmp = dsl_dataset_phys(origin_head)->ds_bp;
3630 		dsl_dataset_phys(origin_head)->ds_bp =
3631 		    dsl_dataset_phys(clone)->ds_bp;
3632 		dsl_dataset_phys(clone)->ds_bp = tmp;
3633 		rrw_exit(&origin_head->ds_bp_rwlock, FTAG);
3634 		rrw_exit(&clone->ds_bp_rwlock, FTAG);
3635 	}
3636 
3637 	/* set dd_*_bytes */
3638 	{
3639 		int64_t dused, dcomp, duncomp;
3640 		uint64_t cdl_used, cdl_comp, cdl_uncomp;
3641 		uint64_t odl_used, odl_comp, odl_uncomp;
3642 
3643 		ASSERT3U(dsl_dir_phys(clone->ds_dir)->
3644 		    dd_used_breakdown[DD_USED_SNAP], ==, 0);
3645 
3646 		dsl_deadlist_space(&clone->ds_deadlist,
3647 		    &cdl_used, &cdl_comp, &cdl_uncomp);
3648 		dsl_deadlist_space(&origin_head->ds_deadlist,
3649 		    &odl_used, &odl_comp, &odl_uncomp);
3650 
3651 		dused = dsl_dataset_phys(clone)->ds_referenced_bytes +
3652 		    cdl_used -
3653 		    (dsl_dataset_phys(origin_head)->ds_referenced_bytes +
3654 		    odl_used);
3655 		dcomp = dsl_dataset_phys(clone)->ds_compressed_bytes +
3656 		    cdl_comp -
3657 		    (dsl_dataset_phys(origin_head)->ds_compressed_bytes +
3658 		    odl_comp);
3659 		duncomp = dsl_dataset_phys(clone)->ds_uncompressed_bytes +
3660 		    cdl_uncomp -
3661 		    (dsl_dataset_phys(origin_head)->ds_uncompressed_bytes +
3662 		    odl_uncomp);
3663 
3664 		dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_HEAD,
3665 		    dused, dcomp, duncomp, tx);
3666 		dsl_dir_diduse_space(clone->ds_dir, DD_USED_HEAD,
3667 		    -dused, -dcomp, -duncomp, tx);
3668 
3669 		/*
3670 		 * The difference in the space used by snapshots is the
3671 		 * difference in snapshot space due to the head's
3672 		 * deadlist (since that's the only thing that's
3673 		 * changing that affects the snapused).
3674 		 */
3675 		dsl_deadlist_space_range(&clone->ds_deadlist,
3676 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3677 		    &cdl_used, &cdl_comp, &cdl_uncomp);
3678 		dsl_deadlist_space_range(&origin_head->ds_deadlist,
3679 		    origin_head->ds_dir->dd_origin_txg, UINT64_MAX,
3680 		    &odl_used, &odl_comp, &odl_uncomp);
3681 		dsl_dir_transfer_space(origin_head->ds_dir, cdl_used - odl_used,
3682 		    DD_USED_HEAD, DD_USED_SNAP, tx);
3683 	}
3684 
3685 	/* swap ds_*_bytes */
3686 	SWITCH64(dsl_dataset_phys(origin_head)->ds_referenced_bytes,
3687 	    dsl_dataset_phys(clone)->ds_referenced_bytes);
3688 	SWITCH64(dsl_dataset_phys(origin_head)->ds_compressed_bytes,
3689 	    dsl_dataset_phys(clone)->ds_compressed_bytes);
3690 	SWITCH64(dsl_dataset_phys(origin_head)->ds_uncompressed_bytes,
3691 	    dsl_dataset_phys(clone)->ds_uncompressed_bytes);
3692 	SWITCH64(dsl_dataset_phys(origin_head)->ds_unique_bytes,
3693 	    dsl_dataset_phys(clone)->ds_unique_bytes);
3694 
3695 	/* apply any parent delta for change in unconsumed refreservation */
3696 	dsl_dir_diduse_space(origin_head->ds_dir, DD_USED_REFRSRV,
3697 	    unused_refres_delta, 0, 0, tx);
3698 
3699 	/*
3700 	 * Swap deadlists.
3701 	 */
3702 	dsl_deadlist_close(&clone->ds_deadlist);
3703 	dsl_deadlist_close(&origin_head->ds_deadlist);
3704 	SWITCH64(dsl_dataset_phys(origin_head)->ds_deadlist_obj,
3705 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
3706 	dsl_deadlist_open(&clone->ds_deadlist, dp->dp_meta_objset,
3707 	    dsl_dataset_phys(clone)->ds_deadlist_obj);
3708 	dsl_deadlist_open(&origin_head->ds_deadlist, dp->dp_meta_objset,
3709 	    dsl_dataset_phys(origin_head)->ds_deadlist_obj);
3710 	dsl_dataset_swap_remap_deadlists(clone, origin_head, tx);
3711 
3712 	dsl_scan_ds_clone_swapped(origin_head, clone, tx);
3713 
3714 	spa_history_log_internal_ds(clone, "clone swap", tx,
3715 	    "parent=%s", origin_head->ds_dir->dd_myname);
3716 }
3717 
3718 /*
3719  * Given a pool name and a dataset object number in that pool,
3720  * return the name of that dataset.
3721  */
3722 int
dsl_dsobj_to_dsname(char * pname,uint64_t obj,char * buf)3723 dsl_dsobj_to_dsname(char *pname, uint64_t obj, char *buf)
3724 {
3725 	dsl_pool_t *dp;
3726 	dsl_dataset_t *ds;
3727 	int error;
3728 
3729 	error = dsl_pool_hold(pname, FTAG, &dp);
3730 	if (error != 0)
3731 		return (error);
3732 
3733 	error = dsl_dataset_hold_obj(dp, obj, FTAG, &ds);
3734 	if (error == 0) {
3735 		dsl_dataset_name(ds, buf);
3736 		dsl_dataset_rele(ds, FTAG);
3737 	}
3738 	dsl_pool_rele(dp, FTAG);
3739 
3740 	return (error);
3741 }
3742 
3743 int
dsl_dataset_check_quota(dsl_dataset_t * ds,boolean_t check_quota,uint64_t asize,uint64_t inflight,uint64_t * used,uint64_t * ref_rsrv)3744 dsl_dataset_check_quota(dsl_dataset_t *ds, boolean_t check_quota,
3745     uint64_t asize, uint64_t inflight, uint64_t *used, uint64_t *ref_rsrv)
3746 {
3747 	int error = 0;
3748 
3749 	ASSERT3S(asize, >, 0);
3750 
3751 	/*
3752 	 * *ref_rsrv is the portion of asize that will come from any
3753 	 * unconsumed refreservation space.
3754 	 */
3755 	*ref_rsrv = 0;
3756 
3757 	mutex_enter(&ds->ds_lock);
3758 	/*
3759 	 * Make a space adjustment for reserved bytes.
3760 	 */
3761 	if (ds->ds_reserved > dsl_dataset_phys(ds)->ds_unique_bytes) {
3762 		ASSERT3U(*used, >=,
3763 		    ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3764 		*used -=
3765 		    (ds->ds_reserved - dsl_dataset_phys(ds)->ds_unique_bytes);
3766 		*ref_rsrv =
3767 		    asize - MIN(asize, parent_delta(ds, asize + inflight));
3768 	}
3769 
3770 	if (!check_quota || ds->ds_quota == 0) {
3771 		mutex_exit(&ds->ds_lock);
3772 		return (0);
3773 	}
3774 	/*
3775 	 * If they are requesting more space, and our current estimate
3776 	 * is over quota, they get to try again unless the actual
3777 	 * on-disk is over quota and there are no pending changes (which
3778 	 * may free up space for us).
3779 	 */
3780 	if (dsl_dataset_phys(ds)->ds_referenced_bytes + inflight >=
3781 	    ds->ds_quota) {
3782 		if (inflight > 0 ||
3783 		    dsl_dataset_phys(ds)->ds_referenced_bytes < ds->ds_quota)
3784 			error = SET_ERROR(ERESTART);
3785 		else
3786 			error = SET_ERROR(EDQUOT);
3787 	}
3788 	mutex_exit(&ds->ds_lock);
3789 
3790 	return (error);
3791 }
3792 
3793 typedef struct dsl_dataset_set_qr_arg {
3794 	const char *ddsqra_name;
3795 	zprop_source_t ddsqra_source;
3796 	uint64_t ddsqra_value;
3797 } dsl_dataset_set_qr_arg_t;
3798 
3799 
3800 /* ARGSUSED */
3801 static int
dsl_dataset_set_refquota_check(void * arg,dmu_tx_t * tx)3802 dsl_dataset_set_refquota_check(void *arg, dmu_tx_t *tx)
3803 {
3804 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3805 	dsl_pool_t *dp = dmu_tx_pool(tx);
3806 	dsl_dataset_t *ds;
3807 	int error;
3808 	uint64_t newval;
3809 
3810 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFQUOTA)
3811 		return (SET_ERROR(ENOTSUP));
3812 
3813 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3814 	if (error != 0)
3815 		return (error);
3816 
3817 	if (ds->ds_is_snapshot) {
3818 		dsl_dataset_rele(ds, FTAG);
3819 		return (SET_ERROR(EINVAL));
3820 	}
3821 
3822 	error = dsl_prop_predict(ds->ds_dir,
3823 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3824 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3825 	if (error != 0) {
3826 		dsl_dataset_rele(ds, FTAG);
3827 		return (error);
3828 	}
3829 
3830 	if (newval == 0) {
3831 		dsl_dataset_rele(ds, FTAG);
3832 		return (0);
3833 	}
3834 
3835 	if (newval < dsl_dataset_phys(ds)->ds_referenced_bytes ||
3836 	    newval < ds->ds_reserved) {
3837 		dsl_dataset_rele(ds, FTAG);
3838 		return (SET_ERROR(ENOSPC));
3839 	}
3840 
3841 	dsl_dataset_rele(ds, FTAG);
3842 	return (0);
3843 }
3844 
3845 static void
dsl_dataset_set_refquota_sync(void * arg,dmu_tx_t * tx)3846 dsl_dataset_set_refquota_sync(void *arg, dmu_tx_t *tx)
3847 {
3848 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3849 	dsl_pool_t *dp = dmu_tx_pool(tx);
3850 	dsl_dataset_t *ds;
3851 	uint64_t newval;
3852 
3853 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3854 
3855 	dsl_prop_set_sync_impl(ds,
3856 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA),
3857 	    ddsqra->ddsqra_source, sizeof (ddsqra->ddsqra_value), 1,
3858 	    &ddsqra->ddsqra_value, tx);
3859 
3860 	VERIFY0(dsl_prop_get_int_ds(ds,
3861 	    zfs_prop_to_name(ZFS_PROP_REFQUOTA), &newval));
3862 
3863 	if (ds->ds_quota != newval) {
3864 		dmu_buf_will_dirty(ds->ds_dbuf, tx);
3865 		ds->ds_quota = newval;
3866 	}
3867 	dsl_dataset_rele(ds, FTAG);
3868 }
3869 
3870 int
dsl_dataset_set_refquota(const char * dsname,zprop_source_t source,uint64_t refquota)3871 dsl_dataset_set_refquota(const char *dsname, zprop_source_t source,
3872     uint64_t refquota)
3873 {
3874 	dsl_dataset_set_qr_arg_t ddsqra;
3875 
3876 	ddsqra.ddsqra_name = dsname;
3877 	ddsqra.ddsqra_source = source;
3878 	ddsqra.ddsqra_value = refquota;
3879 
3880 	return (dsl_sync_task(dsname, dsl_dataset_set_refquota_check,
3881 	    dsl_dataset_set_refquota_sync, &ddsqra, 0,
3882 	    ZFS_SPACE_CHECK_EXTRA_RESERVED));
3883 }
3884 
3885 static int
dsl_dataset_set_refreservation_check(void * arg,dmu_tx_t * tx)3886 dsl_dataset_set_refreservation_check(void *arg, dmu_tx_t *tx)
3887 {
3888 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3889 	dsl_pool_t *dp = dmu_tx_pool(tx);
3890 	dsl_dataset_t *ds;
3891 	int error;
3892 	uint64_t newval, unique;
3893 
3894 	if (spa_version(dp->dp_spa) < SPA_VERSION_REFRESERVATION)
3895 		return (SET_ERROR(ENOTSUP));
3896 
3897 	error = dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds);
3898 	if (error != 0)
3899 		return (error);
3900 
3901 	if (ds->ds_is_snapshot) {
3902 		dsl_dataset_rele(ds, FTAG);
3903 		return (SET_ERROR(EINVAL));
3904 	}
3905 
3906 	error = dsl_prop_predict(ds->ds_dir,
3907 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3908 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, &newval);
3909 	if (error != 0) {
3910 		dsl_dataset_rele(ds, FTAG);
3911 		return (error);
3912 	}
3913 
3914 	/*
3915 	 * If we are doing the preliminary check in open context, the
3916 	 * space estimates may be inaccurate.
3917 	 */
3918 	if (!dmu_tx_is_syncing(tx)) {
3919 		dsl_dataset_rele(ds, FTAG);
3920 		return (0);
3921 	}
3922 
3923 	mutex_enter(&ds->ds_lock);
3924 	if (!DS_UNIQUE_IS_ACCURATE(ds))
3925 		dsl_dataset_recalc_head_uniq(ds);
3926 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3927 	mutex_exit(&ds->ds_lock);
3928 
3929 	if (MAX(unique, newval) > MAX(unique, ds->ds_reserved)) {
3930 		uint64_t delta = MAX(unique, newval) -
3931 		    MAX(unique, ds->ds_reserved);
3932 
3933 		if (delta >
3934 		    dsl_dir_space_available(ds->ds_dir, NULL, 0, B_TRUE) ||
3935 		    (ds->ds_quota > 0 && newval > ds->ds_quota)) {
3936 			dsl_dataset_rele(ds, FTAG);
3937 			return (SET_ERROR(ENOSPC));
3938 		}
3939 	}
3940 
3941 	dsl_dataset_rele(ds, FTAG);
3942 	return (0);
3943 }
3944 
3945 void
dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t * ds,zprop_source_t source,uint64_t value,dmu_tx_t * tx)3946 dsl_dataset_set_refreservation_sync_impl(dsl_dataset_t *ds,
3947     zprop_source_t source, uint64_t value, dmu_tx_t *tx)
3948 {
3949 	uint64_t newval;
3950 	uint64_t unique;
3951 	int64_t delta;
3952 
3953 	dsl_prop_set_sync_impl(ds, zfs_prop_to_name(ZFS_PROP_REFRESERVATION),
3954 	    source, sizeof (value), 1, &value, tx);
3955 
3956 	VERIFY0(dsl_prop_get_int_ds(ds,
3957 	    zfs_prop_to_name(ZFS_PROP_REFRESERVATION), &newval));
3958 
3959 	dmu_buf_will_dirty(ds->ds_dbuf, tx);
3960 	mutex_enter(&ds->ds_dir->dd_lock);
3961 	mutex_enter(&ds->ds_lock);
3962 	ASSERT(DS_UNIQUE_IS_ACCURATE(ds));
3963 	unique = dsl_dataset_phys(ds)->ds_unique_bytes;
3964 	delta = MAX(0, (int64_t)(newval - unique)) -
3965 	    MAX(0, (int64_t)(ds->ds_reserved - unique));
3966 	ds->ds_reserved = newval;
3967 	mutex_exit(&ds->ds_lock);
3968 
3969 	dsl_dir_diduse_space(ds->ds_dir, DD_USED_REFRSRV, delta, 0, 0, tx);
3970 	mutex_exit(&ds->ds_dir->dd_lock);
3971 }
3972 
3973 static void
dsl_dataset_set_refreservation_sync(void * arg,dmu_tx_t * tx)3974 dsl_dataset_set_refreservation_sync(void *arg, dmu_tx_t *tx)
3975 {
3976 	dsl_dataset_set_qr_arg_t *ddsqra = arg;
3977 	dsl_pool_t *dp = dmu_tx_pool(tx);
3978 	dsl_dataset_t *ds;
3979 
3980 	VERIFY0(dsl_dataset_hold(dp, ddsqra->ddsqra_name, FTAG, &ds));
3981 	dsl_dataset_set_refreservation_sync_impl(ds,
3982 	    ddsqra->ddsqra_source, ddsqra->ddsqra_value, tx);
3983 	dsl_dataset_rele(ds, FTAG);
3984 }
3985 
3986 int
dsl_dataset_set_refreservation(const char * dsname,zprop_source_t source,uint64_t refreservation)3987 dsl_dataset_set_refreservation(const char *dsname, zprop_source_t source,
3988     uint64_t refreservation)
3989 {
3990 	dsl_dataset_set_qr_arg_t ddsqra;
3991 
3992 	ddsqra.ddsqra_name = dsname;
3993 	ddsqra.ddsqra_source = source;
3994 	ddsqra.ddsqra_value = refreservation;
3995 
3996 	return (dsl_sync_task(dsname, dsl_dataset_set_refreservation_check,
3997 	    dsl_dataset_set_refreservation_sync, &ddsqra, 0,
3998 	    ZFS_SPACE_CHECK_EXTRA_RESERVED));
3999 }
4000 
4001 /*
4002  * Return (in *usedp) the amount of space written in new that is not
4003  * present in oldsnap.  New may be a snapshot or the head.  Old must be
4004  * a snapshot before new, in new's filesystem (or its origin).  If not then
4005  * fail and return EINVAL.
4006  *
4007  * The written space is calculated by considering two components:  First, we
4008  * ignore any freed space, and calculate the written as new's used space
4009  * minus old's used space.  Next, we add in the amount of space that was freed
4010  * between the two snapshots, thus reducing new's used space relative to old's.
4011  * Specifically, this is the space that was born before old->ds_creation_txg,
4012  * and freed before new (ie. on new's deadlist or a previous deadlist).
4013  *
4014  * space freed                         [---------------------]
4015  * snapshots                       ---O-------O--------O-------O------
4016  *                                         oldsnap            new
4017  */
4018 int
dsl_dataset_space_written(dsl_dataset_t * oldsnap,dsl_dataset_t * new,uint64_t * usedp,uint64_t * compp,uint64_t * uncompp)4019 dsl_dataset_space_written(dsl_dataset_t *oldsnap, dsl_dataset_t *new,
4020     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4021 {
4022 	int err = 0;
4023 	uint64_t snapobj;
4024 	dsl_pool_t *dp = new->ds_dir->dd_pool;
4025 
4026 	ASSERT(dsl_pool_config_held(dp));
4027 
4028 	*usedp = 0;
4029 	*usedp += dsl_dataset_phys(new)->ds_referenced_bytes;
4030 	*usedp -= dsl_dataset_phys(oldsnap)->ds_referenced_bytes;
4031 
4032 	*compp = 0;
4033 	*compp += dsl_dataset_phys(new)->ds_compressed_bytes;
4034 	*compp -= dsl_dataset_phys(oldsnap)->ds_compressed_bytes;
4035 
4036 	*uncompp = 0;
4037 	*uncompp += dsl_dataset_phys(new)->ds_uncompressed_bytes;
4038 	*uncompp -= dsl_dataset_phys(oldsnap)->ds_uncompressed_bytes;
4039 
4040 	snapobj = new->ds_object;
4041 	while (snapobj != oldsnap->ds_object) {
4042 		dsl_dataset_t *snap;
4043 		uint64_t used, comp, uncomp;
4044 
4045 		if (snapobj == new->ds_object) {
4046 			snap = new;
4047 		} else {
4048 			err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &snap);
4049 			if (err != 0)
4050 				break;
4051 		}
4052 
4053 		if (dsl_dataset_phys(snap)->ds_prev_snap_txg ==
4054 		    dsl_dataset_phys(oldsnap)->ds_creation_txg) {
4055 			/*
4056 			 * The blocks in the deadlist can not be born after
4057 			 * ds_prev_snap_txg, so get the whole deadlist space,
4058 			 * which is more efficient (especially for old-format
4059 			 * deadlists).  Unfortunately the deadlist code
4060 			 * doesn't have enough information to make this
4061 			 * optimization itself.
4062 			 */
4063 			dsl_deadlist_space(&snap->ds_deadlist,
4064 			    &used, &comp, &uncomp);
4065 		} else {
4066 			dsl_deadlist_space_range(&snap->ds_deadlist,
4067 			    0, dsl_dataset_phys(oldsnap)->ds_creation_txg,
4068 			    &used, &comp, &uncomp);
4069 		}
4070 		*usedp += used;
4071 		*compp += comp;
4072 		*uncompp += uncomp;
4073 
4074 		/*
4075 		 * If we get to the beginning of the chain of snapshots
4076 		 * (ds_prev_snap_obj == 0) before oldsnap, then oldsnap
4077 		 * was not a snapshot of/before new.
4078 		 */
4079 		snapobj = dsl_dataset_phys(snap)->ds_prev_snap_obj;
4080 		if (snap != new)
4081 			dsl_dataset_rele(snap, FTAG);
4082 		if (snapobj == 0) {
4083 			err = SET_ERROR(EINVAL);
4084 			break;
4085 		}
4086 
4087 	}
4088 	return (err);
4089 }
4090 
4091 /*
4092  * Return (in *usedp) the amount of space that will be reclaimed if firstsnap,
4093  * lastsnap, and all snapshots in between are deleted.
4094  *
4095  * blocks that would be freed            [---------------------------]
4096  * snapshots                       ---O-------O--------O-------O--------O
4097  *                                        firstsnap        lastsnap
4098  *
4099  * This is the set of blocks that were born after the snap before firstsnap,
4100  * (birth > firstsnap->prev_snap_txg) and died before the snap after the
4101  * last snap (ie, is on lastsnap->ds_next->ds_deadlist or an earlier deadlist).
4102  * We calculate this by iterating over the relevant deadlists (from the snap
4103  * after lastsnap, backward to the snap after firstsnap), summing up the
4104  * space on the deadlist that was born after the snap before firstsnap.
4105  */
4106 int
dsl_dataset_space_wouldfree(dsl_dataset_t * firstsnap,dsl_dataset_t * lastsnap,uint64_t * usedp,uint64_t * compp,uint64_t * uncompp)4107 dsl_dataset_space_wouldfree(dsl_dataset_t *firstsnap,
4108     dsl_dataset_t *lastsnap,
4109     uint64_t *usedp, uint64_t *compp, uint64_t *uncompp)
4110 {
4111 	int err = 0;
4112 	uint64_t snapobj;
4113 	dsl_pool_t *dp = firstsnap->ds_dir->dd_pool;
4114 
4115 	ASSERT(firstsnap->ds_is_snapshot);
4116 	ASSERT(lastsnap->ds_is_snapshot);
4117 
4118 	/*
4119 	 * Check that the snapshots are in the same dsl_dir, and firstsnap
4120 	 * is before lastsnap.
4121 	 */
4122 	if (firstsnap->ds_dir != lastsnap->ds_dir ||
4123 	    dsl_dataset_phys(firstsnap)->ds_creation_txg >
4124 	    dsl_dataset_phys(lastsnap)->ds_creation_txg)
4125 		return (SET_ERROR(EINVAL));
4126 
4127 	*usedp = *compp = *uncompp = 0;
4128 
4129 	snapobj = dsl_dataset_phys(lastsnap)->ds_next_snap_obj;
4130 	while (snapobj != firstsnap->ds_object) {
4131 		dsl_dataset_t *ds;
4132 		uint64_t used, comp, uncomp;
4133 
4134 		err = dsl_dataset_hold_obj(dp, snapobj, FTAG, &ds);
4135 		if (err != 0)
4136 			break;
4137 
4138 		dsl_deadlist_space_range(&ds->ds_deadlist,
4139 		    dsl_dataset_phys(firstsnap)->ds_prev_snap_txg, UINT64_MAX,
4140 		    &used, &comp, &uncomp);
4141 		*usedp += used;
4142 		*compp += comp;
4143 		*uncompp += uncomp;
4144 
4145 		snapobj = dsl_dataset_phys(ds)->ds_prev_snap_obj;
4146 		ASSERT3U(snapobj, !=, 0);
4147 		dsl_dataset_rele(ds, FTAG);
4148 	}
4149 	return (err);
4150 }
4151 
4152 /*
4153  * Return TRUE if 'earlier' is an earlier snapshot in 'later's timeline.
4154  * For example, they could both be snapshots of the same filesystem, and
4155  * 'earlier' is before 'later'.  Or 'earlier' could be the origin of
4156  * 'later's filesystem.  Or 'earlier' could be an older snapshot in the origin's
4157  * filesystem.  Or 'earlier' could be the origin's origin.
4158  *
4159  * If non-zero, earlier_txg is used instead of earlier's ds_creation_txg.
4160  */
4161 boolean_t
dsl_dataset_is_before(dsl_dataset_t * later,dsl_dataset_t * earlier,uint64_t earlier_txg)4162 dsl_dataset_is_before(dsl_dataset_t *later, dsl_dataset_t *earlier,
4163     uint64_t earlier_txg)
4164 {
4165 	dsl_pool_t *dp = later->ds_dir->dd_pool;
4166 	int error;
4167 	boolean_t ret;
4168 
4169 	ASSERT(dsl_pool_config_held(dp));
4170 	ASSERT(earlier->ds_is_snapshot || earlier_txg != 0);
4171 
4172 	if (earlier_txg == 0)
4173 		earlier_txg = dsl_dataset_phys(earlier)->ds_creation_txg;
4174 
4175 	if (later->ds_is_snapshot &&
4176 	    earlier_txg >= dsl_dataset_phys(later)->ds_creation_txg)
4177 		return (B_FALSE);
4178 
4179 	if (later->ds_dir == earlier->ds_dir)
4180 		return (B_TRUE);
4181 	if (!dsl_dir_is_clone(later->ds_dir))
4182 		return (B_FALSE);
4183 
4184 	if (dsl_dir_phys(later->ds_dir)->dd_origin_obj == earlier->ds_object)
4185 		return (B_TRUE);
4186 	dsl_dataset_t *origin;
4187 	error = dsl_dataset_hold_obj(dp,
4188 	    dsl_dir_phys(later->ds_dir)->dd_origin_obj, FTAG, &origin);
4189 	if (error != 0)
4190 		return (B_FALSE);
4191 	ret = dsl_dataset_is_before(origin, earlier, earlier_txg);
4192 	dsl_dataset_rele(origin, FTAG);
4193 	return (ret);
4194 }
4195 
4196 void
dsl_dataset_zapify(dsl_dataset_t * ds,dmu_tx_t * tx)4197 dsl_dataset_zapify(dsl_dataset_t *ds, dmu_tx_t *tx)
4198 {
4199 	objset_t *mos = ds->ds_dir->dd_pool->dp_meta_objset;
4200 	dmu_object_zapify(mos, ds->ds_object, DMU_OT_DSL_DATASET, tx);
4201 }
4202 
4203 boolean_t
dsl_dataset_is_zapified(dsl_dataset_t * ds)4204 dsl_dataset_is_zapified(dsl_dataset_t *ds)
4205 {
4206 	dmu_object_info_t doi;
4207 
4208 	dmu_object_info_from_db(ds->ds_dbuf, &doi);
4209 	return (doi.doi_type == DMU_OTN_ZAP_METADATA);
4210 }
4211 
4212 boolean_t
dsl_dataset_has_resume_receive_state(dsl_dataset_t * ds)4213 dsl_dataset_has_resume_receive_state(dsl_dataset_t *ds)
4214 {
4215 	return (dsl_dataset_is_zapified(ds) &&
4216 	    zap_contains(ds->ds_dir->dd_pool->dp_meta_objset,
4217 	    ds->ds_object, DS_FIELD_RESUME_TOGUID) == 0);
4218 }
4219 
4220 uint64_t
dsl_dataset_get_remap_deadlist_object(dsl_dataset_t * ds)4221 dsl_dataset_get_remap_deadlist_object(dsl_dataset_t *ds)
4222 {
4223 	uint64_t remap_deadlist_obj;
4224 	int err;
4225 
4226 	if (!dsl_dataset_is_zapified(ds))
4227 		return (0);
4228 
4229 	err = zap_lookup(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4230 	    DS_FIELD_REMAP_DEADLIST, sizeof (remap_deadlist_obj), 1,
4231 	    &remap_deadlist_obj);
4232 
4233 	if (err != 0) {
4234 		VERIFY3S(err, ==, ENOENT);
4235 		return (0);
4236 	}
4237 
4238 	ASSERT(remap_deadlist_obj != 0);
4239 	return (remap_deadlist_obj);
4240 }
4241 
4242 boolean_t
dsl_dataset_remap_deadlist_exists(dsl_dataset_t * ds)4243 dsl_dataset_remap_deadlist_exists(dsl_dataset_t *ds)
4244 {
4245 	EQUIV(dsl_deadlist_is_open(&ds->ds_remap_deadlist),
4246 	    dsl_dataset_get_remap_deadlist_object(ds) != 0);
4247 	return (dsl_deadlist_is_open(&ds->ds_remap_deadlist));
4248 }
4249 
4250 static void
dsl_dataset_set_remap_deadlist_object(dsl_dataset_t * ds,uint64_t obj,dmu_tx_t * tx)4251 dsl_dataset_set_remap_deadlist_object(dsl_dataset_t *ds, uint64_t obj,
4252     dmu_tx_t *tx)
4253 {
4254 	ASSERT(obj != 0);
4255 	dsl_dataset_zapify(ds, tx);
4256 	VERIFY0(zap_add(ds->ds_dir->dd_pool->dp_meta_objset, ds->ds_object,
4257 	    DS_FIELD_REMAP_DEADLIST, sizeof (obj), 1, &obj, tx));
4258 }
4259 
4260 static void
dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t * ds,dmu_tx_t * tx)4261 dsl_dataset_unset_remap_deadlist_object(dsl_dataset_t *ds, dmu_tx_t *tx)
4262 {
4263 	VERIFY0(zap_remove(ds->ds_dir->dd_pool->dp_meta_objset,
4264 	    ds->ds_object, DS_FIELD_REMAP_DEADLIST, tx));
4265 }
4266 
4267 void
dsl_dataset_destroy_remap_deadlist(dsl_dataset_t * ds,dmu_tx_t * tx)4268 dsl_dataset_destroy_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4269 {
4270 	uint64_t remap_deadlist_object;
4271 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4272 
4273 	ASSERT(dmu_tx_is_syncing(tx));
4274 	ASSERT(dsl_dataset_remap_deadlist_exists(ds));
4275 
4276 	remap_deadlist_object = ds->ds_remap_deadlist.dl_object;
4277 	dsl_deadlist_close(&ds->ds_remap_deadlist);
4278 	dsl_deadlist_free(spa_meta_objset(spa), remap_deadlist_object, tx);
4279 	dsl_dataset_unset_remap_deadlist_object(ds, tx);
4280 	spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4281 }
4282 
4283 void
dsl_dataset_create_remap_deadlist(dsl_dataset_t * ds,dmu_tx_t * tx)4284 dsl_dataset_create_remap_deadlist(dsl_dataset_t *ds, dmu_tx_t *tx)
4285 {
4286 	uint64_t remap_deadlist_obj;
4287 	spa_t *spa = ds->ds_dir->dd_pool->dp_spa;
4288 
4289 	ASSERT(dmu_tx_is_syncing(tx));
4290 	ASSERT(MUTEX_HELD(&ds->ds_remap_deadlist_lock));
4291 	/*
4292 	 * Currently we only create remap deadlists when there are indirect
4293 	 * vdevs with referenced mappings.
4294 	 */
4295 	ASSERT(spa_feature_is_active(spa, SPA_FEATURE_DEVICE_REMOVAL));
4296 
4297 	remap_deadlist_obj = dsl_deadlist_clone(
4298 	    &ds->ds_deadlist, UINT64_MAX,
4299 	    dsl_dataset_phys(ds)->ds_prev_snap_obj, tx);
4300 	dsl_dataset_set_remap_deadlist_object(ds,
4301 	    remap_deadlist_obj, tx);
4302 	dsl_deadlist_open(&ds->ds_remap_deadlist, spa_meta_objset(spa),
4303 	    remap_deadlist_obj);
4304 	spa_feature_incr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
4305 }
4306